2017年6月25日~29日にかけて，カナダのバンクーバーにて開催されましたThe 23rd Annual Meeting of the Organization for Human Brain Mappingに参加いたしました．
 
この学会は，神経イメージングの知識，経験を分かち合い，最新の研究と今後の展望についての情報交換，議論の場となることを目的に開催されています．M2の片山，石原，玉城，和田，吉武，M1の相本，藤井聖香，池田，三好，水野，石田翔也，中村圭佑の総勢15名が参加しました．26日に日和先生，片山(M2)，藤井聖香(M1)，中村圭佑(M1)，27日に石原(M2)，相本(M1)，三好(M1)，石田翔也(M1)，28日に玉城(M2)，水野(M1)，29日に和田(M2)，吉武(M2)，萩原(M2)，池田(M1)がポスターセッションで発表いたしました．

 
発表題目は以下の通りです．

<!>
私にとっては2度目の国際学会であり，前回とは学会の規模，研究室からの参加数などが異なり，前回とは違った経験をすることができたと感じています．私はポスター発表が最終日であったため，前日までの研究室メンバーの発表姿をみることができ，それによりモチベーションをより上げることができました．発表では練習してきた説明できましたが，質問を聴き，その質問に答えることは難しく，とにかく黙ることなく質問を聴き返したり，自分が考えたことを話す努力をしてコミュニケーションを途切れずに心がけました．去年参加した国際学会より，多くの研究を知り，多くの方とお話しできたので良い体験ができました．前回の国際学会に比べて，ポスター発表を積極的に聴き質問することができた点が成長できたと感じています．しかし，自分の研究と関係する少しのキーワードに絞ってしかポスター発表を聴いていなかったので，多くの発表がある学会であったためより広範な研究に関して発表を聴けたらよかったと思いました． OHBMに参加することで，研究室の中では知る機会が少ない，ヒトの脳に関する研究の全体的な動向を知ることができたのがよかったと思っています．今回得た知識や不足していると感じている部分を踏まえ，更に研究に励んでいきたいと思います.


【文責：M2 萩原】
学会参加報告書

報告者氏名	水野めぐみ
発表論文タイトル
発表論文英タイトル	Functional connectivity analysis of brain activity during cooperative behavior using fNIRS
著者	水野めぐみ,日和悟，廣安知之
主催	医療情報システム研究室
講演会名	23nd Annual Meeting of the Organization of Human Brain Mapping
会場	Vancouver Convention Centre
開催日程	2017/06/25-2017/06/29

 
 

1. 講演会の詳細

2017/06/25から2017/06/29にかけて，Vancouver Convention Centreにて開催されました23nd Annual Meeting of the Organization of Human Brain Mappingに参加いたしました．この学会は，人間の脳の解剖学的・機能的組織、および健康や病気について理解を進めるため，人間の脳組織に関する研究に携わっている様々な背景の研究者が集い，研究者間のコミュニケーションを促進し、人間の脳組織における教育を促進することを目的としている。
25日-29日の全日参加いたしました．本研究室からは他に廣安先生，日和先生，石原さん，和田さん，玉城さん，片山さん，萩原さん，吉武さん，池田，藤井，三好，相本，石田（翔），中村（圭）が参加しました．

2. 研究発表
   • 発表概要

私は28日の午後のPoster Secssion「Imaging Methods：fNIRS」にて発表いたしました．発表の形式はポスター発表で，2時間自由に発表および質疑応答を行う時間となっておりました．
今回の発表は，「Functional connectivity analysis of brain activity during cooperative behavior using fNIRS」です．以下に抄録を記載致します．

【Introduction】 Humans typically belong to the organizations, such as schools or companies, and often live with others. Cooperation is essential for a group to work well together; understanding the intention of others and acting accordingly is also required. Important brain functions for social life can be clarified by examining the brain during cooperative behaviors. Cooperative behavior is related to imitating and synchronizing behavior among individuals [1]. In this study, we investigated brain functions that synchronize behaviors with a stimulus whose presentation interval changes proportionally over time. 【Methods】 Twenty healthy subjects (10 females, aged 21-23 years) participated in this experiment. The synchronized tapping task was used to investigate timing control functions [2]. Brain activity during tapping synchronized to a sound stimulus with a presentation interval that increases proportionally over time was measured using fNIRS. Brain functional connectivity using temporal correlations in cerebral blood flow changes was analyzed using a graph theoretical analysis. The network threshold was set to preserve an edge density of 15%, and the degree centrality, which is the total number of links in each region, was calculated as the network characteristic. Subjects were divided into two groups (A = 12 subjects, B = 8 subjects) by hierarchical clustering using the Ward method after 116-channel degree centralities of all subjects were decomposed into 11-dimensional values using a principal component analysis. The tendency of each group was examined from two points: the regions with high degree and the difference in response time to the stimulus. 【Results】 Fig. 1 shows the brain regions associated with high degree centrality for each group. In group A, the degree centrality of the left middle frontal gyrus (LMFG) and the triangular part of inferior frontal gyrus (TrIFG) was high. The average response time in group A was slower than the stimulus presentation. In group B, the degree centrality was high in the left middle frontal gyrus (LMFG), left middle occipital gyrus (LMOG), left postcentral gyrus (LPoG), left supramarginal gyrus (LSMG), and right middle temporal gyrus (RMTG). The average response time in group B was faster than the stimulus presentation. The dorsolateral prefrontal cortex (DLPFC), including the middle frontal gyrus (MFG), is related to sustained attention [3]. Moreover, the middle temporal gyrus (RMTG) was reported to be involved in the information processing of motion [4]. Predicted action involves a top-down process [5]; therefore, the results for group A suggest that the left frontal region plays a central role in controlling the attention to the stimulus, which facilitates rapid responses for synchronization. In group B, this network includes not only the left frontal region but also the parietal and temporal regions. We speculate that these regions adjust timing by predicting the next stimulus in addition to attention control. 【Conclusions】 We examined brain activity during cooperative behavior. Functional brain networks during a synchronized tapping task were investigated based on the temporal correlation of cerebral blood flow measured by fNIRS. These data suggest that the important brain regions of the functional network for cooperative behavior are the left middle frontal gyrus (LMFG), left middle occipital gyrus (LMOG), left postcentral gyrus (LPoG), left supramarginal gyrus (LSMG), and right middle temporal gyrus (RMTG).

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
 
・質問内容1
明治大学の都地さんからの質問です． なぜグラフ理論を用いているのか。この質問に対し、計測CHが116CHあり、ネットワークがとても複雑であるため、その複雑なネットワークの構造を理解するために、グラフ理論を用いることが必要であると回答しました。
 
・質問内容2
質問者の氏名を控え損ねてしまいました．
Betweenness centrality などいろいろ中心性の指標がある中でなぜDegree centralityを検討したのか。この質問に対して、多くの部位と協調している部位同士を繋ぐ中心的なネットワーク（上層）を検討するのではなく、多くの部位と協調している部位のネットワーク（下層）を見ることが重要であると考えているからですと回答しました。この学会を通して、今後はrichclubネットワークの階層構造を検討する必要があることを学びました。
 
・質問内容3
質問者の氏名を控え損ねてしまいました．
バンドパスフィルタの範囲に関して、0.33Hzは検討し直す必要があるとご指摘頂きました。文献調査をしている中でも0.1Hzのものが多く検討し直す必要があると考えています。
 
・質問内容4
質問者の氏名を控え損ねてしまいました．
行動データの結果について、クラスタ内の被験者は応答のプラスマイナスの傾向が共通しているのか。この質問に対して、平均の応答なのでクラスタ内でもマイナスの場合とプラスの場合がありますと回答しました。今回の発表では行動データに対して質問されることが何度かありました。
 

• 感想

初めての国際学会参加でした。国内学会とは違い英語力が乏しく、研究の議論を深めることはできませんでした。しかし、これまで英語で対話するという経験がなかったので、よい経験になりました。次回、国際学会に参加するときには、もっと英語で意見交換ができるように、準備をしっかりし、英語力をつけて対話ができるように思いました。

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました。

発表タイトル       ：　Differential contributions of transient and sustained channels across the visual hierarchy 著者                  ： Catie Chang, NIH, Bethesda, MD, セッション名       ： Perception and Attention Abstruct            ： 【Introduction】 Prevailing psychophysical models propose that the human visual system contains separate temporal channels for processing transient and sustained visual stimuli (Kulikowski & Tolhurst, 1973; McKee & Taylor, 1984; Watson, 1985). While neural responses in primary visual cortex (V1) are consistent with a two temporal channel model (Horiguchi et al., 2009), the relative contribution of the two channels and their functional significance at later stages of the visual hierarchy is unknown. 【Methods】 To address this gap in knowledge, we scanned 12 subjects in a 3T scanner using a novel fMRI paradigm that estimates independent transient and sustained contributions to blood oxygen level dependent (BOLD) responses across visual cortex using three experiments. All experiments used the same phase-scrambled stimuli, trial durations, and fixation task, and only varied in the temporal presentation of stimuli. Experiment 1 was designed to strongly activate the sustained channel by presenting a single static image continuously for trials lasting 2, 4, 8, 15 or 30 s. Experiment 2 was designed to strongly activate the transient channel by presenting in each 2, 4, 8, 15 or 30 s trial 30 different images, each shown for 33 ms and followed by a blank screen lasting 33-967 ms. Experiment 3 was designed to drive responses in both sustained and transient channels by presenting in each 2, 4, 8, 15 or 30 s trial 30 images for 67-1000 ms in a continuous fashion without intervening blanks. To model BOLD responses we implemented a neural model with two temporal channels: (1) a sustained channel of ongoing neural responses over the duration of a stimulus, and (2) a transient channel that generates a brief neural response at the onset and offset of an image. Then, we convolved the neural response of each channel with a HRF to predict BOLD responses. We estimated the contributions (β weights) of the sustained and transient channels using data from Experiments 1 and 2 and cross-validated the model by quantifying how well it predicts responses for independent data from Experiment 3. 【Results】 The standard model of BOLD responses depends only on stimulus duration, consequently predicting higher responses in Experiment 1 than in Experiment 2, and similar responses in trials of the same duration across Experiments 1 and 3. Different from these predictions: (1) in short trials (<8 s) responses in Experiment 2 were higher than Experiment 1 and (2) across all trial durations responses in Experiment 3 were higher than Experiment 1. These data suggest that transient responses that are not considered by the standard model contribute to BOLD responses. Indeed, a neural model with two temporal channels (sustained and transient) predicts BOLD responses significantly better than the standard model (Figure 1a, main effect of model, F(2, 22) = 56.49, p < 0.001). This improvement is larger at later stages of the visual hierarchy than V1 (Figure 1b, model-by-area interaction, F(2, 22) = 6.45, p < 0.01). Additionally, we find differential contributions of the sustained and transient neural channels across the visual hierarchy. Early visual areas (V1, V2, V3) have significant contributions from both channels, ventral regions (hV4, VO1, VO2) have twice as much contribution from the transient than sustained channel, and lateral regions (LO1, LO2, hMT+) are driven primarily by the transient channel with minimal sustained contributions (Figure 1c). 【Conclusions】 These results demonstrate that the two temporal channel model accurately predicts responses to stimuli with a broad range of temporal characteristics and furthermore reveals functional differences across stages of the visual hierarchy. Critically, these data suggest that any model of BOLD responses in any cortical system needs to consider the temporal dynamics of the experimental paradigm to accurately predict brain responses.

MISLでも、刺激に対してモデルを作っているので、着目した。持続的な神経応答と一時的な神経応答の２つの刺激を考慮してモデルが考えられており、興味深い内容だった。
 

発表タイトル       ：　Meta-analysis of heterogeneous EEG studies using hierarchical event descriptor (HED) tags 著者                  ： Nima Bigdely Shamlo, Alejandro Ojeda1, Tim Mullen, Kay Robbins セッション名       ： Poster Session Abstruct            ： 【Introduction】 Analyzing EEG trials across studies with different experimental paradigms requires a formally-defined common conceptual space. Hierarchical event descriptor (HED) tags (Bigdely-Shamlo, Cockfield, et al., 2016) provides such a common space by providing an extensible controlled vocabulary to describe study trial (event) and state details. Here we present results of the first EEG (meta)analysis on over half a million trials from studies with different paradigms and investigate the statistical validity of using HED tags across these studies. 【Methods】 We used data from six studies in ESS Level 2 (Bigdely-Shamlo, Makeig, & Robbins, 2016) containers. The containers included HED strings associated with study events along with other information needed for automated analysis of the data such as channel labels, montage, and tasks assigned to each data recording. Data in ESS Level 2 containers is also preprocessed using PREP pipeline (Bigdely-Shamlo, Mullen, Kothe, Su, & Robbins, 2015). EEG data was then band-passed 0.5-20 Hz. Single-trial EEG activity epochs from all scalp channels were then extracted for all events in each session of each study (total of 634,359 epochs) along with HED strings associated with each trial. We then computed two types of similarity matrices to be used in RSA analysis. (a) between single trial pairs of each two event types. (b) for each predefined HED tag group, between the HED strings associated with each two event types. For (a), data from all channels in each trial epoch was vectorized and the average Pearson correlation of all pairs of events across the two event types was computed (see Fig. 1, Left). The values in the resulting event similarity matrix g(Ei, Ej) were then normalized as described in the figure. To form HED-based event similarity matrices (b), we first formed tag groups for the validation analysis by creating a list of all HED tags from all 118 event types. For each tag, all parent levels were also included in the list, e.g. for the tag Sensory presentation/Visual/Rendering type/Screen/2d, the parent tags (1) Sensory presentation/Visual/Rendering type/Screen (2) Sensory presentation/Visual were also included. Tags that contained only a single child were removed from the list. This was to prevent unfounded generalization, i.e. if only instances of Sensory Item/Symbolic/Character/Letter are present in our data (but not any instances of the parent tags), there is not enough data supporting the analysis of the more general parent tags, e.g. Item/Symbolic/. Analysis tag groups were then formed by applying connected-component labeling (Dillencourt, Samet, & Tamminen, 1992) to thresholded (>0.9) Pearson correlation of tag matches and event types. Finally, we performed Representational Similarity Analysis (RSA) (Kriegeskorte, Mur, & Bandettini, 2008) between EEG-based event similarity matrix (a) and HED-based event similarity matrices, each associated with an analysis tag group. 【Results】 Table 1 shows the RSA results. Tag groups that are found to be associated with statistically significant similarities (after correcting for multiple comparisons with FDR (Benjamini & Hochberg, 1995), p<0.05) across EEG dynamics in our data are marked with *. 【Conclusions】 Our results show that at least for a subset of HED tags, events tagged similarly across studies are more similar to each other. This is a necessary condition for HED tags to be useful in EEG meta-analysis and contributes to the validation of HED tagging. More detailed validation of HED hierarchy (or any other ontology for EEG events) can be conducted by scaling up our analysis method on a larger number of diverse, tagged EEG studies.

EEGデータを用いたメタアナリシス解析による研究。月例発表会にて、メタアナリシスというものを初めて知ったが世界でも注目されていることを実感した。
 
 

発表タイトル       ：　Neural Responses to Dynamic Pain Expression of Same-Race and Other-Race Faces 著者                  ： Wenxin Li, Shihui Han セッション名       ： Poster session Abstruct            ： 【Introduction】 Previous neuroimaging studies of empathy employed faces with static pain and neutral expressions and have shown stronger frontal activity as early as 120 ms after sensory stimulation to pain (vs. neutral) expression of same-race than other-race faces (Sheng and Han, 2012). It remains unknown how the brain responds to dynamic pain expression of same-race and other-race faces. Here we tested the hypothesis that brain activity is more sensitive to dynamic changes of facial expression from neutral to pain of same-race than other-race faces by recording event-related brain potentials (ERPs) from Chinese female adults while they performed judgments of Asian and Caucasian faces. 【Methods】 We recruited 32 Chinese female adults (mean = 21.3 years, SD = 2.5 years). Participants performed a pain judgment task on two sets of face stimuli that were created by morphing a neutral face (0%) and a painful face (100%) of the same model (16 Asian and 16 Caucasian models were used) with a step of 10%. Each face was presented for 80 ms and followed by a fixation cross with a duration varying randomly between 600ms and 1200ms. EEG was recorded during 10 blocks of 352 trials (each face was presented once in a random order). For each participant, two Asian models (1 male) and two Caucasian models (1 male) were assigned as a target by adding a white frame for pain judgments (painful vs. neutral). Participants responded to each target by pressing one of two keys. Behavioral responses were fit by a Weibull function, p=1-e^(-(x/α)^β ), where p is the percentages of perceived pain expression, x is the percentage of pain pixels in the morphed faces, α is the point of subjective equality (PSE), and β is the slope of the sigmoidal response function. EEG data analysis focused on the P2 amplitude at 140-190 ms to non-target stimuli over frontocentral electrodes. ANOVAs were conducted on P2 amplitudes with Race (Asian vs. Caucasian) and Percentage of pain pixels (0% to 100%) as within-subjects variables. 【Results】 The analysis of behavioral data did not show significant difference in the PSE between same-race and other-race faces. However, ANOVAs of the P2 amplitudes showed significant main effects of race (Fz: F(1,31)=99.29,p< .001; FCz: F(1,31)=113.49,p< .001; Cz: F(1,31)=124.93,p< .001) and percentage of pain-pixel (Fz: F(10,310)=9.58,p< .001; FCz: F(10,310)=10.40,p<.001; Cz: F(10,310)=11.45,p<.001). The P2 amplitudes were greater to other-race than same-race faces and increased as a function of the percentage of pain-pixels in photos. To estimate the sensitivity of brain activity to dynamic pain expression of same-race and other-race faces, we constructed a classical linear regression model (P2=ax+b) of the relationship between the P2 amplitudes and the percentage of pain-pixel for each participant. A paired t-test revealed a significantly greater slope of the linear function (i.e., a) for same-race than other-race faces (t(33)=2.22, p=0.034), further supporting that the P2 amplitudes were more sensitive to variation of painful information in same-race compared with other-race faces. 【Conclusions】 Even the behavioral data failed to uncover any difference in subjective sensitivity to variation of painful information in same-race and other-race faces, our ERP results provide evidence that the brain activity in an early time window is more sensitive to dynamic changes of facial expression from neutral to pain of same-race than other-race faces.

共感の研究。画像呈示実験はよくあるが、呈示画像が動的であるところが面白いと感じた。動画では定量的にそのときの刺激状態を定義することはできないが、呈示画像を動的にすることで、評価もしやすくかつ動的な「表情」に対する神経応答を見ることができる。
 

発表タイトル       ：　Neural underpinnings of mutual gaze and joint attention using hyperscanning functional MRI 著者                  ： Hiroki　Tanabe セッション名       ： Brain-to-brain synchrony early in life: What can we learn from different hyperscanning techniques? Abstruct            ： Mutual gaze provides a communicative link between humans, prompting joint attention, which is the ability to coordinate attention between interactive social partners with respect to objects or events to share an awareness of them. Joint attention is of particular importance during early social development representing the prerequisite of theory-of mind and social communication. To elucidate their neural underpinnings, we conducted several experiments employing hyperscanning functional MRI combined with online video cameras and voice exchange system. I will show the results of these studies and discuss core neural mechanisms of mutual gaze and joint attention.

MRIを2台使用し、カメラ越しに相手が見えるように環境が構築されており、実験設備がとても整っていた。また、実験課題は3つのコンディションがあり、実験自体の構成も複雑だった。今後実験設計を考える参考にしていこうと思う。Hyperscannigの研究の講演を聞くことができて、今後の研究のモチベーションが上がった。聴講だけでは理解できなかったことも多かったが、文献を調査し研究に活かしていきたい。
 

発表タイトル       ：　Exploring the neural evidence of mother-infant entrainment: Inter-brain synchronized hemodynamic activity 著者                  ： Yasuyo Minagawa セッション名       ： Brain-to-brain synchrony early in life: What can we learn from different hyperscanning techniques? Abstruct            ： In the present study, we performed functional Near-Infrared Spectroscopy (fNIRS) based hyperscanning to examine synchronized brain activity in mothers and 3- to 4-month-old infants. Cerebral hemodynamic changes in mother-infant dyads were measured under three conditions: i.e. (1) breast feeding, (2) resting state during mother holding her infant and (3) resting state during separation (control). The results showed inter-brain synchronized hemodynamic activity in the prefrontal cortex which was significantly larger in the holding condition than in the control condition. This may reflect a neural correlate of mother-infant bonding.

赤ちゃんの計測を実際に行っている研究の話を聞くことができて面白かった。協調性を考える上で協調性を獲得する発達過程の赤ちゃんの研究は関係が深いと思う。赤ちゃんと大人はHRFが異なることなど興味深い内容だった。しかし、赤ちゃんを計測するのは本当に問題がないのかと少し疑問に思ってしまう。
 
参考文献

• 2017 ANNUAL MEETING, https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageID=3744

学会参加報告書

報告者氏名	萩原里奈
発表論文タイトル	暗算課題中に得られた機能データを用いた脳状態の分類
発表論文英タイトル	Classification of brain states using functional data obtained during a mental arithmetic task
著者	萩原里奈，日和悟，廣安知之
主催	Organization for Human Brain Mapping
講演会名	The 23rd Annual Meeting of the Organization for Human Brain Mapping
会場	Vancouver Convention Centre
開催日程	2017/6/25-2017/6/29

 
 

1. 講演会の詳細

2017/6/25-29に，Vancouver Convention Centreにて開催されましたThe 23rd Annual Meeting of the Organization for Human Brain Mappingに参加いたしました．この大会は，ヒトの脳マッピングに関心のある研究者・医師・学生が参加します．神経イメージングの知識，経験を分かち合い，最新の研究と今後の展望についての情報交換，議論の場となることを目的に開催されています．
本研究室からは他に廣安先生，日和先生，M2の片山さん，吉武さん，石原さん，和田さん，玉城さん，M1の水野さん，藤井さん，池田さん，石田翔也さん，三好さん，相本さん，中村圭佑さんが参加しました．
 

2. 研究発表
   • 発表概要

私は29日のポスターセッションに参加いたしました．発表の形式はポスター発表で，2時間自由に参加者の方と議論を行いました．
今回の発表は，「Classification of brain states using functional data obtained during a mental arithmetic task」と題して発表いたしました．以下に抄録を記載します．

【Introduction】 Working memory (WM), which is a temporary storage system that processes information, is necessary for daily life [1,2]. A previous study on the functional connectivity (FC) revealed that WM is a complex system and each brain region cooperates by exchanging information [3]. However, there are a paucity of studies on individual brain states due to differences in the strategies for WM tasks. Therefore, it is necessary to assess brain states due to difference in performance of WM. In this study, we compared brain states that were classified using FC data.【Materials & Methods】 Thirty-two healthy adults (average age: 22.0 ± 1.2) while performing a mental arithmetic task was scanned using fMRI. The task consisted of mental arithmetic with a low WM load (Low-WM task) or a high WM load (High-WM task). Preprocessing was applied to acquired fMRI images using SPM8. The whole brain was partitioned into 116 regions of interest (ROI) based on an automated anatomical labeling atlas, and a correlation matrix was created by a ROI-to-ROI analysis using the Conn toolbox [4]. This matrix was adjusted to a connection density of 15% by thresholding. The Jansen-Shannon divergence, which indicates the similarity between subjects’ matrix, was calculated between each matrix, and a hierarchical clustering analysis was performed [5,6]. The activated regions in each classified group were compared using t-tests. Moreover, each matrix was binarized, and the degree (Deg) and clustering coefficient (CC) were calculated using the Brain Connectivity Toolbox [7]. The Deg is the number of connections with other regions, and the CC indicates the ratio of neighboring regions to each other. 【Results】Subjects were classified into two groups (Fig. 1), and the average correct answer rate of the High-WM task was 56.0 ± 1.23 for Group A (n = 13) and 50.2 ± 19.8 for Group B (n = 19). The superior parietal lobule (SPL) in Group A and the precuneus, lingual gyrus, calcarine sulcus, precentral gyrus, middle occipital lobe, and middle frontal gyrus in Group B were extracted as regions whose activations were significantly higher for High-WM task when compared with the Low-WM task. These regions were responsible for WM functions in previous studies [3,8]. Group B appeared to require many active regions, whereas Group A needed fewer activated regions to complete the task. Moreover, the Deg of the superior temporal gyrus (STG) and cerebellar vermis as well as the CC of the cerebellum (CRBL) and cerebellar vermis were significantly higher for the High-WM task than the Low-WM task (p < .05). The CRBL supports the execution of WM by internal utterance, and the STG is involved in auditory short-term memory [8,9]. Thus, our result suggests that internal speech is repeated during the WM task, and that these regions act as hubs. On the other hand, the Deg of the cuneus, superior occipital lobe, SPL, CRBL, and cerebellar vermis as well as the CC of the middle occipital lobe, fusiform gyrus, precuneus, and CRBL in Group B were significantly higher during the High-WM task when compared with the Low-WM task (p < .05). The SPL is involved in updating information [10]. Thus, it appears that Group B strengthened the cooperation of the SPL with other regions to update information. Our connectivity analysis showed that Group A, which had a high average score, had few regions with a strong connection to other regions, whereas Group B, which had a low average performance, increased the number of regions with strengthened connectivity. 【Conclusion】Trends among the groups classified from the FC data during a WM task were examined in this study. Subjects underwent fMRI while performing mental arithmetic tasks with different degrees of difficulty. The participants were classified into two groups. One group had a low score with activity in multiple regions, whereas the other group had a high score with limited activated areas. Therefore, FC data can be used to identify distinct brain states during WM tasks.

 

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
 
・質問内容1
質問者の氏名を控え損ねてしまいました．こちらの質問はクラスタリングが上手くでき，このクラスタリング方法が有用であることを示すaccuracyなどの結果はあるのかというものでした．今回はそのような解析は行えておらず，今後の課題であると回答致しました．
 
・質問内容2
質問者の氏名を控え損ねてしまいました．こちらの質問はJSダイバージェンスとは何かというものでした．こちらの質問に対する回答ですが，相関行列間の類似度を分布として捉えていると回答致しました．
 
・質問内容3
　質問者の氏名を控え損ねてしまいました．こちらの質問は，脳領域のdifficultタスクとeasyタスクのdegreeの違いの検討に関して，統計的な解析を行っているのかというものでした．この質問に対して，各群のN数が少ないため統計的な解析は行っておらず，difficultタスクとeasyタスクのdegree値の差が大きいものを抽出したと回答致しました．
 
・質問内容4
　質問者の氏名を控え損ねてしまいました．こちらの質問は，図に示される脳のネットワークはどのように抽出したのかというものでした．この質問に対して，各被験者の相関行列をエッジ密度15%で閾値処理し，各群において抽出されたdegreeのタスク間の差が大きい脳領域との結合を被験者ごとにネットワークとして抽出したと回答致しました．
 
・質問内容5
　質問者の氏名を控え損ねてしまいました．こちらの質問は，なぜこのように被験者群を分類したのかというものでした．階層的クラスタリングの結果から，それぞれの主な枝ごとにそのグループがどのような脳状態になっているのかを検討しようと思い，今回の被験者分類を行ったと回答致しました．
 

• 感想

私にとっては2度目の国際学会であり，前回とは学会の規模，研究室からの参加数などが異なり，前回とは違った経験をすることができたと感じています．ポスター発表は発表日がそれぞれ異なり，発表時に研究室のメンバーが発表の応援に来てくれたため緊張をほぐして発表することができました．また，私はポスター発表が最終日であったため，前日までの研究室メンバーの発表姿をみることができ，それによりモチベーションをより上げることができたと感じています．発表では練習してきた説明はできるのですが，質問を聴き，その質問に答えるということはやはり難しく，とにかく黙ることなく質問を聴き返したり，自分が考えたことを話す努力をしてコミュニケーションを途切れさせないようにできたのがよかったと思います．また，学会規模が前回参加した国際学会より大きかったことから，多くの研究を知ることができ，多くの方とお話しできたので良い体験ができました．前回の国際学会に比べて，ポスター発表を積極的に聴き質問することができた点が成長できたと感じています．しかし，自分の研究と関係する少しのキーワードに絞ってしかポスター発表を聴いていなかったので，多くの発表がある学会であったためより広範な研究に関して発表を聴けたらよかったと思いました．研究室の中では知ることができる研究の範囲が狭くなりがちですが，OHBMに参加することでヒトの脳に関する研究の全体的な動向を知ることができたのがよかったと思っています．今回得た知識や不足していると感じている部分を踏まえ，更に研究に励んでいきたいと思います．
 

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました．
 

発表タイトル       ：　Cortical-striatal connectivity in obsessive compulsive disorder is hyper-modulated by working memory 著者                  ： Jane Harness セッション名       ： Poster Session Abstruct            ： Introduction: Obsessive compulsive disorder (OCD) is a disorder that causes obsessive and intrusive thoughts presumed to drive repetitive compulsions. As most behavior is sub served by synchrony in cortical, striatal and thalamic sub-networks, it is likely that repetitive behaviors in OCD arise from dysfunctional interactions between these regions (Nakao et al., 2009). These interactions are presumed to be hyper-functional in nature, either compensating for latent connective deficits or reflect a hyper-synchrony that resists the modulatory effects of cognitive control, a domain that is impaired in the disorder (Diwadkar et al., 2015). Here we evaluated the ability of basic working memory tasks to a) modulate functional connectivity (FC) between brain networks in OCD and healthy controls (HC) and b) to estimate significant differences between groups on sub-network pairs identified in a). We demonstrate that OCD are characterized by significantly increased FC in dACC-centric networks between 1) the right and left basal ganglia and 2) other frontal and cortical regions. Methods: Thirty-five participants (13 male, 22 female, mean age=16.3 yrs.) with a diagnosis of OCD and 35 controls (18 male, 17 female, mean age=17 yrs.) participated in the fMRI study (Siemens Verio 3T) of the verbal n-back task (Casey et al.,1995) used to drive activity in frontal, cingulate, striatal, parietal and thalamic networks. Data were processed in SPM8 using typical methods. From first-level statistical models, effects of interest contrasts were used to identify task-related voxels for each participant and were forwarded for second level analyses in which the one-way analysis of variance used Group (OCD, HC) as the single factor of interest. In this second level model, our intent was to identify brain regions that were co-activated across groups, using conjunction analysis (Nichols et al., 2005). Significant peaks from each co-activated region of interest were identified (pvoxel<0.05). These peaks were submitted to subsequent undirected FC analyses, assessed using typical methods (Whitfield-Gabrieli & Nieto-Castanon, 2012; Silverstein et al., 2016). Analyses proceeded in a two-step process: First sub-networks that were significantly modulated by task (pFDR<0.05) in both OCD and HC were identified. Next, these co-modulated sub-network pairs were submitted for subsequent analyses of inter-group differences (OCD ≠ HC). Results: Figure 1 shows sub-networks that were significantly modulated by task (pFDR<0.05) in both OCD and HC for 0 back (1a) and 1 back (1b). From these co-modulated sub-network pairs, Figure 2 depicts inter-group differences (OCD > HC; p<0.05). Figure 2a represents these pairs for 0 back and figure 2b for 1 back. There were no sub-network pairs with HC>OCD.

強迫性障害（OCD）郡とコントロール群の0-backと1-backの機能的コネクティビティの比較に関する発表でした．各群の機能的ネットワークを作成し，群間を検定することで群に特有のネットワークを抽出していました．0-backと1-backでは負荷にそれほど違いがないように感じていましたが，1-backの結合が多い結果となっていました．2-backに関してもデータはあるが解析をしていないと言っており，2-backにおける結果も気になりました．
 

発表タイトル       ：　Working memory in childhood onset schizophrenia patients and their nonpsychotic siblings 著者                  ： Siyuan Liu セッション名       ：  Poster Session Abstruct            ： Introduction: Childhood onset schizophrenia (COS), defined as onset of psychosis before age 13, is a rare and severe form of the disorder (Nicolson and Rapoport 1999). Working memory (WM) deficits are consistently reported in the adult-onset schizophrenia (AOS) literature (Goldman-Rakic 1994), and abnormal brain activations (Minzenberg, et al. 2009) and functional connectivity (Dauvermann, et al. 2014) have been found in adult-onset patients. COS patients and their nonpsychotic siblings show WM impairments as well (Gochman, et al. 2004; Karatekin and Asarnow 1998), but it is unclear whether similar abnormal functional activations and connections underlie these processes to those seen in adult-onset peers. Here, we conducted an fMRI experiment of n-back paradigm to examine whether COS patients and their siblings show impairment in WM function compared with healthy controls, and examine whether their WM related brain activation and connectivity patterns are abnormal. Methods: 32 COS patients (21.3+-6.1 years), 30 nonpsychotic siblings (19.4+-4.1), and 39 healthy controls (20.0+-4.5), matched for age and sex, were scanned at 3T and performed a block-designed paradigm. It included 4 permuted runs, one for each of 1-, 2-back letter (verbal) and location (visual) WM tasks, and 0-back was embedded in each run as the baseline. The total duration was 17.64 min. COS patients met DSM-IV criteria for schizophrenia with the onset of psychosis before 13. An ICA based pipeline was used to preprocess fMRI images (Xu, et al. 2014). DVARS, a measure of artifact-induced changes in image intensity, was matched across three groups (<0.87%, p=0.27). In addition to typical activation analyses, we also evaluated the averaged functional connectivity, the average Pearson correlation of each voxel with all others (Cole, et al. 2010). Random-effects ANOVA models were used to draw statistical inferences at the group level. A family-wise error of 0.05 was used to determine corrected significance. Results: COS patients scored significantly lower in accuracy rate than controls in all tasks (Cohen’s d > 1, p < .0001). To be noted, only 40% COS patients were able to accomplish 2-back tasks, indicating that they suffer a severe loss of WM function. Unlike patients, siblings showed no significantly lower accuracy rates compared to controls. However, when switching from 1- to 2-back tasks, their averaged effect sizes increased from 0.1 to 0.4 and p value reached 0.16, indicating that siblings suffer a subthreshold WM functional loss. fMRI analyses revealed a similar brain activation pattern in controls to that of adult controls (Owen, et al. 2005), including robust activations in bilateral dorsolateral and ventrolateral prefrontal cortex (DLPFC and VLPFC); medial and lateral posterior parietal cortex; dorsal cingulate and medial premotor cortex (ACC and preSMA); cerebellum; and thalamus as well as caudate nuclei. Compared with controls, both patients and siblings showed hypo activations in the frontal parietal network, ACC and caudate. Furthermore, compared with controls, only COS patients showed significantly reduced functional connectivity from the above hypo-activated regions to other areas during tasks, and siblings did not. Conclusions: Only a low percentage of COS patients able to accomplish the 2-back tasks and their large effect sizes of all tasks reveal that COS patients suffer a severe loss of WM functions. Our imaging findings identified dysfunction of the frontal-parietal network in COS patients, resembling the pattern seen in most studies of adult-onset patients. Both behavioral and imaging results show siblings to be intermediate between the patients and controls, which is aligned with the previous findings in twins of adult-onset patients (Karlsgodt, et al. 2007), supporting that abnormal brain activations could be a trait marker.

本発表は，小児発症統合失調症（COS）と感覚統合機能不全（SID）患者とコントロール群における文字と位置の1back，2backを成績，脳活性，機能的コネクティビティを比較した研究でした．COS群とコントロール群，SID群とコントロール群の成績の有意な違いを比較しワーキングメモリ能力の違いを示し，その上で脳活性の違い，脳活性をした領域をSeedとした結合の違いを検討しており，各結果を1つずつ検証した上で結論に導いていることを感じ，自分の研究をする際にも参考にしたいと感じました．また，n-back課題に関しても研究室で行われている単純なものではなく，条件が違うものを複数使っていることが興味深いと思いました．
 

発表タイトル       ：　Single-patient analysis of impaired RS-fMRI connectivity 著者                  ： Azzurra Invernizzi セッション名       ： Poster Session Abstruct            ： Introduction: Resting-state fMRI (rs-fMRI) has been largely used to investigate intrinsic functional connectivity in the human brain, occurring in the absence of any stimulus. Several studies have proven rs-fMRI useful in detecting functional connectivity alterations, even in earlier stage, of neurodevelopmental disorders, aging and psychiatric conditions [1]. Although a wide range of methods have been developed to analyze rs-fMRI data, a complete pipeline to perform a single-patient level analysis is still lacking. Based on CONN functional connectivity toolbox [2] and SPM12, we developed an optimized single-patient data analysis pipeline of rs-fMRI data. Methods: The pipeline is essentially based on three main steps: optimized preprocessing of rs-fMRI data, ROI-to-ROI functional connectivity analysis and linear statistical approach (Fig.1). First, rs-fMRI data are segmented, then spatially realigned, coregistered, normalized to the standard MNI-152 template. Deformation fields are computed on white matter (WM), grey matter (GM) and cerebral spinal fluid (CSF) of the structural MR image, in order to obtain a total deformation field that maps the MR image to MNI space. This total deformation is applied to the bias corrected rs-fMRI data, which are then resampled to 3mm isotropic voxels. Next, whole-brain ROI-to-ROI functional connectivity analyses are performed using the 96 ROIs based on Harvard-Oxford Atlas. Head realignment parameters, WM and CSF signals are regressed from the data following the CompCor-strategy [3] implemented in the CONN toolbox. Residual time-series of the rs-fMRI images are then band-pass filtered (0,0009<f<0,08 Hz). No global signal regression is applied. For each subject, we extract the mean time series by averaging across all voxels in each ROI. We then compute the bivariate correlation coefficients for each pair of ROIs. The resultant ROI-to-ROI correlation values are Fisher z-transformed. Finally, single-case statistical analysis is run on the ROI-to-ROI correlation matrices. Specifically, a t-test [4] was computed between the complete control dataset and each single patient. To evaluate the performance of our pipeline, we used rs-fMRI data from right-handed male volunteers included in the Autism Brain Imaging Data Exchange I and II (ABIDE I and II) [5]. Inclusion criteria for control and ASD subjects were: (i) age between 5 and 36 years, which represents approximately two standard deviations from the overall mean age across all male participants (14.7 ± 6.2 years); (ii) a mean frame-wise displacement (FD) lower than 0.5 mm [6] and (iii) with successful preprocessing. These criteria turned out in a rs-fMRI dataset composed of 390 controls and 95 ASD males. Results: We performed a cluster analysis, based on the functional connectivity matrices obtained in the single-patient statistical analysis, to define 3 main patient phenotypes. For each of these groups of patients, we evaluated the following scores from the Autism Diagnostic Interview (ADI): reciprocal social interaction, verbal, non-verbal and restricted patterns of behavior. Accordingly, we identified three main different patient profiles that present similar behavioral impairments and functional connectivity data (Fig. 2).

CONNとSPM12に基づく，単一患者データ解析パイプラインの開発に関する発表でした．機能的コネクティビティマトリックスを用いてクラスタ分析を行い，3つの患者表現を定義し，行動パターンを用いてそれらを評価していました．解析ソフトや解析方法が似ている部分から興味を持ちましたが，クラスタ分析以降の解析に関して十分に理解することが出来なかったのが残念でした．
 

発表タイトル       ：　Brain Network of Emotion Regulation in Soldiers with Trauma 著者                  ： D Rangaprakash セッション名       ： ORAL SESSION: Emotion and Motivation Abstruct            ： Introduction: Our ability to shape our emotional experience is termed emotion regulation (ER) [1], involving voluntary modification of emotions elicited in response to exogenous stimuli. Several functional MRI activation studies have consistently identified the middle frontal gyrus (MFG), anterior cingulate and insula to be involved in it [1]. Their limitation lies in the inability to explain the interrelationship between these regions, i.e. connectivity. The brain network of ER either in healthy adults or in psychiatric disorders like posttraumatic stress disorder (PTSD) and mild-traumatic brain injury (mTBI) has been elusive. Emotion dysregulation (ED) is regarded a primary cause for many symptoms observed in PTSD and mTBI [2]. Using fMRI data collected during an ER task, we obtained the network of ER in healthy soldiers and ED in soldiers with comorbid PTSD and post-concussion syndrome (PCS, or chronic mTBI). Methods: 59 male U.S. Army soldiers were recruited (comorbid PCS+PTSD=36, combat controls=37, matched in age, race and education). FMRI data was acquired in a Siemens Verio 3T scanner (EPI sequence, TR/TE=600/30ms, flip-angle=55o, voxel size=3.5×3.5×5mm3). The ER task (Fig.1) was similar to Urry et.al. [3]. Participants were presented images eliciting a negative emotional response, and were asked to either “maintain” their emotional response, or “suppress” it (reduce negative feelings, requiring ER). There were 4 task blocks, with 24 trials in each block. Standard pre-processing was performed in SPM (realignment, smoothing [8mm kernel], normalization to MNI space). We first identified significantly activated regions during ER (see Fig.2 for region selection procedure). Hemodynamic deconvolution was performed [4] on mean time series extracted from identified regions, to minimize the non-neural intra-subject HRF variability [5]. We employed effective connectivity (EC) modeling using Granger causality (GC) [6] to assess directional causal relationships between identified regions, similar to recent works [7]. Subject-wise EC between all regions were obtained, using which the networks of ER in healthy soldiers (suppress>maintain) and its impairment in PCS+PTSD (control>PCS+PTSD for ‘suppress’ condition) were obtained (p<0.001, Bonferroni corrected) (Fig.3). We provide novel evidence for the brain networks of both ER and ED in a clinical population. Results: We investigated brain networks of ER in healthy soldiers, and ED in PCS+PTSD. We defined our ROIs around the 9 regions activated during the ER task (Fig.4). With EC analysis, we found the ER network having a top-down structure with the MFG driving the rest of the network (insula, medial prefrontal, amygdala and lateral parietal regions) (Figs 5a,5b,5c). During ED this network was imbalanced with reduced prefrontal connectivity and elevated subcortical and lateral parietal connectivity (Figs 5d,5e,5f). Our ER network fits well with prior findings [1, 8], which identified the pivotal role of MFG in the initiation of ER. MFG is implicated in executive functions like cognitive control [1], which are necessary for regulating emotions. Soldiers with PTSD exhibit impaired emotional processing [9] and impaired cognitive functions associated with the MFG [10],. All directional connections are traceable to the MFG, implying that MFG could be the source of ER [1]. As for ED, the MFG emerged as the key source of disruption in PCS+PTSD. All connections from MFG had reduced connectivity, whose “ripple-effect” culminated in disinhibition of amygdala, which might translate to symptoms like flashbacks, trauma re-experiencing and hyperarousal. This fits well with behavioral manifestations of these conditions [2]. Conclusions: In summary, we identified the MFG as pivotal to ER in healthy soldiers and ED in PCS+PTSD. Our findings are significant given that these regions are implicated in prior activation studies [1, 8], but a precise understanding of the underlying network structure and their causal relationships had not emerged so far.

兵士を被験者として，感情調整のネットワークを検討する発表でした．賦活解析，メタアナリシス，動的ネットワーク解析など様々な解析を行っており，興味深い研究でした．また，賦活解析で抽出された脳領域を中心とした，ネットワーク構造を検討しており，今後の自身の研究の参考にしたいと思いました．
 

発表タイトル       ：　Towards mapping the neural substrates of the residual variance in human working memory 著者                  ： Christelle van Antwerpen セッション名       ： ORAL SESSION: Higher Cognitive Functions Abstruct            ： Introduction: Traditionally working memory has been viewed as having a limited storage capacity, able to hold only a small, fixed number of items {1,2}. More recently working memory function has been conceptualised as being constrained by both short-term memory storage and processing efficiency {3,4,5}. However, there is evidence that a residual variance in working memory performance that is not wholly explained by these storage and processing components exists. This residual variance is the least understood component of working memory and there is much debate over what this variance represents, particularly over whether or not it is executive in nature {6,7,8,9}. We aimed to examine the neural mechanisms that underlie working memory, and shed further light on each of the subcomponents of working memory performance. We used fMRI to address the conceptualisation of working memory, specifically related to the residual variance and what it represents. Methods: Participants: 62 healthy volunteers (49 females) were recruited (mean age = 34.05 years, age range 16-66) Paradigm: We used five tasks modified from Bayliss et al., (2003)3. A simple processing task with no storage requirements. Two complex memory tasks required performance of a concurrent processing task: a complex verbal memory task and a complex spatial memory task. Two tasks were simple verbal and spatial storage tasks (without a concurrent processing task). Data acquisition: 3T Siemens Skyra scanner, 32 channel head coil. fMRI scans were T2*-weighted gradient EPI images, TR of 2.5s, 3mm ³ voxels, 40 slices covering the whole brain. A T1-weighted inversion recovery MPRAGE giving 1mm³ voxels, 192 slices was also acquired. Analysis: performed in SPM8. A general linear model was used to estimate brain regions showing greater task specific activation versus rest. Five t-contrasts: processing, storage, verbal working memory, spatial working memory and complex working memory. The verbal working memory, spatial working memory and complex working memory contrasts represent the residual variance, however separating verbal and spatial working memory contrasts allows us to directly test the domain generality of working memory. Consistent effects were then tested with one-sample t-test for each contrast and a two-sample t-test to compare between verbal a spatial domains was conducted. Significant results threshold p<0.05 Family-wise error corrected. Results: Accuracy results showed no significant effects for both task domain and task difficulty. The reaction time results showed a significant increase in complex tasks than storage tasks. In complex task reaction times were significantly faster in the spatial compared to verbal domain. The fMRI results showed that during the processing component, the bilateral superior temporal gyrus shows strong increases in the BOLD signal. In the analysis of the storage component results indicate strong activation of the bilateral inferior parietal lobule as well as activations in cerebellum (culmen). The analysis of residual variance revealed consistent activations in the bilateral temporal parietal junction, anterior cingulate cortex and prefrontal cortex. A direct comparison between spatial and verbal complex working memory showed no differences in the pattern of activation, suggesting the same neural pathway underlies both verbal and spatial working memory domain. Conclusions: Our study is the first to use fMRI to address the debate of the residual variance and what it may represent [6,7,8,9]. Our findings show that the residual variance is supported by neural networks known to support attention and executive function and strongly indicates that the residual variance represents an executive component in working memory. Furthermore our findings demonstrate a common neural pathway shared by both verbal and spatial domains of working memory. These findings indicates that the residual is domain-general. Together these findings strongly indicate that the residual variance is indeed executive.

ワーキングメモリの神経メカニズムを調べ，各サブコンポーネントを明らかにする発表でした．2種類のタスクを用いて，行動データと共に脳活動の解析が行われ，各タスクを比較することで検討したい脳活動を明らかにしていました．こちらの発表においても，1つ1つのことを明らかにし，それぞれの結果を合わせて結論に導いていると感じました．
 
参考文献

• The 23rd Annual Meeting of the Organization for Human Brain Mapping,

https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageID=3734
学会参加報告書

報告者氏名	池田幸樹
発表論文タイトル	視覚刺激課題におけるHRFに適応させたGLM解析の評価
発表論文英タイトル	Evaluation of a GLM analysis with adaptive hemodynamic response function on a visual stimulus task
著者	池田幸樹，日和悟，廣安知之
主催	医療情報システム研究室
講演会名	23rd Annual Meeting of the Organization of Human Brain Mapping
会場	Vancouver, Canada
開催日程	2017/06/25-2017/06/29

 
 

1. 講演会の詳細

2017/06/25から2017/06/29にかけて，Vancouver, Canadaにて開催されました23rd Annual Meeting of the Organization of Human Brain Mapping（OHBM2017）に参加いたしました．このOHBMは，科学者間のコミュニケーションを促進し，健康・病気の人間の脳における解剖学的・機能的組織の理解を進めることを目的として開催されています．
私は全日程参加し，本研究室からは他に廣安先生，日和先生，石原さん，玉城さん，和田さん，片山さん，萩原さん，吉武さん，相本さん，三好さん，石田(翔)，中村(圭)さん，藤井(聖)さん，水野さんが参加されました．
 

2. 研究発表
   • 発表概要

私は29日の午後のポスターセッションに参加いたしました．発表の形式はポスター発表で，貼り出し期間は2日間，セッション時間は2時間となっておりました．
今回の発表について以下に抄録を記載致します．

前書き：一般線形モデル（GLM）に基づく脳活動の解析法は、fMRIにおいて一般的に用いられているが、この方法は機能的近赤外分光法（fNIRS）にも容易に適用できる。 GLMでは、血行動態関数（HRF）とボクセル関数との畳み込みによってシミュレートされた脳血流変化が生成され、活性化の程度は、測定された脳活動データに適合させることによって推定される[1] [2]。通常、HRFは所定の形状に固定されており、boxcar関数の値は1に固定されているが、これらは刺激および個人によって異なると仮定する[3] [4]。この問題を解決するために、著者らは、実際のデータとシミュレートされたデータとの間の誤差を最小にするために、HRFとboxcar関数のパラメータを最適化する新しいGLM法を提案した。本論文では、提案手法の有効性を視覚刺激課題の分析データにより検証した。メソッド：提案された方法では、HRFの3つのパラメータ、第1のピーク遅延tp、第2のピーク遅延時間tu、第1のピークと第2のピークの比A、および刺激提示時のboxcar関数の値（ここでは、刺激ベクトルと呼ぶ）は、 HRFの畳み込みによって得られたシミュレートされた血流変化とボックスカー関数と測定データとの間の誤差は最小限にされる。しかし、本論文では、HRFパラメータを共通の設定p = 6、u = 10、A = 6に固定し、刺激ベクトルのみを最適化した。提案した方法の有効性を評価するための実験データとして、健常者（n = 20;女性10名;年齢22 +/- 0.1歳）の脳血流変化データを、116チャンネルで測定したNAPSデータセットfNIRS（日立、ETG-7100）を用いた。 fNIRSの全ての測定チャネルは、自動化解剖学的標識（ALL）に基づく脳領域と関連していた。タスク区間の積分値は、提案手法により得られたシミュレートされた脳血流変化データから算出した。積分値の最大値を有する上位5つの脳領域を活性化領域とみなした。結果：図1は提案手法で処理したfNIRSデータから抽出した活性化領域を示す。左右の背側上前頭回、右前頭回、左後頭部回の4領域を抽出した。特に、不快な感情の認知制御に関連しているため、左背側上前頭回が活性化されていると考えられている[5]。これらの結果から、提案手法はタスクを扱う領域を抽出できることが示された。 結論：本論文では、不快な画像を見ながら脳活動のfNIRSデータにHRFとboxcar関数の値を最適化する新しいGLM解析を適用し、その有効性を検証した。その結果、不快な感情に関連する脳領域を抽出することができ、本手法の有効性が確認された。

 

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
 
・質問内容1
こちらの質問は，なぜ最適化するのかというものでした．この質問に対して，従来のGLM解析ではタスクで一定の刺激が出ていると仮定していましたが，実際はチャンネルや個人によって異なるためと回答しました．
 
・質問内容2
こちらの質問は，他のタスクで最適化を行ったらどうなるのかというものでした．この質問に対して，今後他の実験にも適応していく予定であると回答しました．
 
・質問内容3
こちらの質問は，なぜレストでも刺激が出ているのかというものでした．この質問に対して，レストでも十時固視点を提示しておりタスクからの影響を受け刺激が出ている可能性があるためだと回答しました．
 

• 感想

本学会に参加し，一番印象に残っていることは私の拙い英語の発表を聞いた後，多くの人に興味深い面白い研究だと言っていただいたことです． 今までfNIRSの新しいGLM解析は本当に正しいのかなど様々な思いを抱えて研究を進めてきましたが，初めての外部発表で今まで進めてきた研究に自信を持つことができました．また，英語力の向上と自分の研究への理解を深める必要性を実感しました．研究への理解が低いため，質問されてることが理解できても自信を持って答えることができない場面が多くありました．この経験を生かし，今後はきちんと理解して研究を進めていこうと思いました．
 

3. 聴講

今回の学会では，下記の5件の発表を聴講しました．
 

発表タイトル    ：　Meta-analysis of heterogeneous EEG studies using hierarchical event descriptor (HED) tags 著者                ： Nima Bigdely Shamlo, Alejandro Ojeda, Tim Mullen, Kay Robbins セッション名      ： Poster Session Abstruct       ：異なる実験パラダイムを用いた研究を通したEEG試験の分析には，正式に定義された共通概念空間が必要である．階層的事象記述子（HED）タグは，試験試行（事象）および状態の詳細を記述するための拡張可能な制御語彙を提供することによって，このような共通の空間を提供する．本研究では，異なるパラダイムを用いた研究からの50万回以上の試行についての最初のEEG（メタ）分析の結果を提示し，これらの研究でHEDタグを使用する統計的妥当性を調べる． ESS Level 2の容器で6つの研究のデータを使用しました。コンテナには、調査イベントに関連するHEDストリングと、チャンネルラベル、モンタージュ、各データ記録に割り当てられたタスクなどのデータの自動解析に必要なその他の情報が含まれていました。 ESSレベル2コンテナのデータもPREPパイプラインを使用して事前処理されます。脳波データを次に0.5~20Hzでバンド通過させた。その後、各試験の各セッション（合計634,359エポック）のすべての事象について、各試験に関連するHEDストリングと共に、すべての頭皮チャンネルからの単一試験EEG活動時期を抽出した。次に、RSA解析に使用する2種類の類似性行列を計算した。 （a）各2つのイベントタイプの1つの試行ペアの間。 （b）各所定のHEDタググループについて、各2つのイベントタイプに関連するHEDストリングの間にある。 （a）では、各試行エポックの全チャンネルからのデータがベクトル化され、2つのイベントタイプにわたるすべてのイベント対の平均ピアソン相関が計算された（図1、左を参照）。得られたイベント類似性行列g（Ei、Ej）の値を、図に記載されているように正規化した。HEDベースのイベント類似性マトリックス（b）を形成するために、118個のすべてのイベントタイプからすべてのHEDタグのリストを作成することにより、検証分析のためのタググループを最初に形成した。各タグについて、すべての親レベルもリストに含まれていた。感覚提示/視覚/レンダリングタイプ/スクリーン/ 2D、親タグ（1）感覚提示/視覚/レンダリングタイプ/スクリーン（2）感覚提示/視覚も含まれていた。単一の子のみを含むタグはリストから削除した。これは根拠のない一般化を防ぐためであり、つまり、感覚アイテム/記号/文字/文字のインスタンスのみがデータに存在する場合（親タグのインスタンスではない場合）、より一般的な親タグの分析をサポートするデータが不十分である．例えばアイテム/シンボリック/．次に、タグマッチおよびイベントタイプの閾値（> 0.9）ピアソン相関に連結成分ラベリングを適用することによって分析タググループを形成した。最後に、EEGベースのイベント類似性マトリックス（a）と分析タググループに関連付けられたHEDベースのイベント類似性マトリックスとの間のRepresentational Similarity Analysis（RSA）を実施した。 表1にRSAの結果を示します。我々のデータのEEG動態の間に、統計的に有意な類似性（FDRとの複数の比較を補正した後（Benjamini＆Hochberg、1995）、p <0.05）と関連していることが判明しているタグ群には*印が付いている。 我々の結果は、少なくともHEDタグのサブセットについては、研究の間に同様にタグ付けされた事象が互いに類似していることを示している。これは、HEGタグがEEGメタ分析に有用であることを必要条件とし、HEDタグ付けの検証に寄与する。 HED階層（またはEEGイベントのための他のオントロジー）のより詳細な検証は、より多くの多様なタグ付きEEG研究について分析方法を拡大することによって行うことができる。この研究は陸軍研究所によって支援され、協定番号W911NF-10-2-0022の下で達成された。

この発表で着目した点は，EEGを用いてメタアナリシス解析を行っていることです．fMRIではメタアナリシス解析が広くおこなわれていますが，EEGでは珍しいとのことです．Methodがフローチャートになっていてわかりやすかったです．
 

発表タイトル    ：Differential contributions of transient and sustained channels across the visual hierarchy 著者                ： Anthony Stigliani1, Brianna Jeska1, Kalanit Grill-Spector1 セッション名      ： Oral Session Abstruct       ：優勢な精神物理学的モデルは、人間の視覚システムが、一時的および持続的な視覚刺激を処理するために別々の時間チャネルを含むことを提案する。第1次視覚野（V1）の神経応答は2つの時間チャネルモデルと一致しているが、2つのチャネルの相対的な寄与と視覚的階層の後期段階におけるそれらの機能的意義は不明である。 知識のこのギャップに対処するために、3つの実験を用いて視覚野全体の血中酸素レベル依存性（BOLD）応答への独立した一過性および持続的な寄与を推定する新規fMRIパラダイムを用いて、3Tスキャナで12人の被験者をスキャンした。すべての実験は、同じ位相スクランブル刺激、試行期間、および固定タスクを使用し、刺激の時間的提示においてのみ変化した。実験1は、2,4,8,15または30秒持続する試験のために単一の静止画像を連続的に提示することによって持続的なチャネルを強く活性化するように設計された。実験2は、各2,4,8,15または30秒の試行で33ミリ秒間表示された30個の異なる画像を提示し、続いて33-967ミリ秒間続くブランク画面を提示することによって過渡チャネルを強く活性化するように設計された。実験3は、各2,4,8,15、または30秒間の試行30の画像を67-1000msの間、連続的に空白を介在させずに提示することによって、持続的および過渡的な両方のチャネルで応答を駆動するように設計された。 BOLD応答をモデル化するために、我々は、（1）刺激の持続時間にわたる進行中の神経応答のチャネル、および（2）発症時およびオフセット時に短い神経応答を生成する一時的チャネルの2つの時間チャネルを有する神経モデルを実施した。画像。次に、BOLD応答を予測するために、各チャネルの神経応答をHRFで畳み込んだ。実験1および2のデータを使用して、持続的チャネルおよび過渡的チャネルの寄与（β加重）を推定し、実験3の独立データに対する応答をどれだけうまく予測するかを定量化してモデルを交差検証した。 BOLD反応の標準モデルは、刺激持続時間のみに依存し、その結果、実験1よりも実験2よりも高い応答を予測し、実験1および3の同じ期間の試験で同様の反応を予測する。これらの予測とは異なった。実験2および実験3の応答が実験1よりも高かった。これらのデータは、標準モデルでは考慮されていない過渡応答がBOLD応答に寄与していることを示唆している。実際、2つの時系列チャネル（持続的および一時的）を有する神経モデルは、標準モデルよりもBOLD応答を有意に良好に予測する（図1a、モデルの主な効果、F（2,22）= 56.49、p <0.001）。この改善は、V1よりも視覚的階層の後の段階でより大きい（図1b、モデルごとの相互作用、F（2,22）= 6.45、p <0.01）。さらに、我々は、視覚的な階層全体にわたる持続的および一時的な神経チャネルの異なる寄与を見出す。初期の視覚領域（V1、V2、V3）は両方のチャネルから有意な寄与を有し、腹側領域（hV4、VO1、VO2）は持続チャネルより2倍の寄与を有し、横領域（LO1、LO2、hMT +）主に一時的な寄与を最小限に抑えた過渡チャネルによるものである（図1c）。 これらの結果は、2つの時間的チャネルモデルが、広い範囲の時間特性を有する刺激に対する応答を正確に予測し、さらに、視覚的階層の段階全体にわたる機能的差異を明らかにすることを示す。重要なことに、これらのデータは、脳の応答を正確に予測するために、実験的パラダイムの時間的ダイナミクスを考慮する必要があることを示唆している。

この発表で着目したのは一過性の視覚刺激と持続性の視覚刺激を比較しているところです．視覚野の反応は視覚刺激の長さと数に依存していることを示していました．今後の視覚刺激実験の設計を考える際，参考にしてみたいです．
 

発表タイトル    ：Data-driven estimates of vigilance are linked with fluctuations in task performance 著者                ： Catie Chang1, Jacco de Zwart1, Hendrik Mandelkow1, Jeff Duyn1 セッション名      ： Oral Session Abstruct       ：警戒における自発的な変動（脳の興奮）は，認知および行動と密接に相互作用し，fMRIにおける被験者間および被験者内変動の主要な原因となる可能性がある．古典的な警戒の指標（EEGやアイ・ビデオなど）は必ずしも入手可能ではなく，入手が困難な場合があるため，fMRIデータ自体から警戒の変動を推測する方法は大きな価値がある．マカクでの最近の研究では，フレームごとにfMRIからの警戒変動（覚醒と軽い睡眠の間のドリフト）を追跡するアプローチが示されている．ここでは，先述のアプローチを人間の被験者に適用し，この手順から推定された警戒レベルが，継続的な作業記憶タスク中の被験者のパフォーマンスの変動を予測するものであるかどうかを調べる． 1つのコホート（「タスクコホート」，N = 9,3T，TR = 2秒，12分）は，1回の作業記憶タスクを行い，205回の試行（それぞれ1文字で構成される）インターバル（ISI）は3.5秒である．不正確な回答の試行は省略した．残りの患者の反応時間（RT）がRT範囲の上位5％および下位5％に収まった試験は，被験者ごとに「遅い」または「高速」のカテゴリーにそれぞれ割り当てられた．同時にEEG-fMRI（「脳波コホート」，N = 7，眼閉鎖休止状態，3T，TR = 1.5秒）を受けた第2群の被験者では，警戒レベルの変化に関連する空間パターンを導出した”テンプレート”）．このテンプレートは，各TRにおけるEEGアルファ（8-12Hz）対シータ（3-7Hz）パワーの比として定義される警戒回帰因子をfMRIデータと相関させることによって得られた．得られたマップは，MNI空間に整列され，脳波コホートの被験者間で平均化された（図1）．すべてのfMRIデータ（両方のコホート）は，スライスタイミング補正，運動コアギストレーション，線形および二次トレンドの除去，生理的アーチファクトの減少，およびMNI空間へのアライメントで前処理された．次に，このテンプレートをタスクコホートのfMRIデータに適用した．具体的には，各走査について，テンプレートマップと各連続するfMRI容積（すなわち，各TR）との間の空間相関を計算することによって，推定警戒レベル（EVL）を表す時間経過が導出される． 1-backタスクの各試行について，先行する2.5秒間隔（5秒の血行力学的遅延を仮定すると，与えられた試行後2.5秒-5秒の間隔内でEVLを平均化することに相当する）にわたって平均EVLを照会し， 1つの被験者は，RT全体の変動性が小さいため（<100ms），省略された． （i）警戒テンプレート：脳波コホート（図1）から得られたテンプレートマップは，警戒と休息状態のfMRI活動との相関の以前に説明されたパターンと一致する．具体的には，警戒の増加は，多くの皮質領域にわたる活動の広範な減少と関連し，視床を含む領域の重大な増加を伴う．（ii）EVLと作業成績との関係：作業コホートの8人の被験者のうち6人について，fMRIデータのみから推定された警戒レベルは，より遅い反応時間と比較してより早い試行で高かった（図2） ．残りの2人の被験者のEVLは，より早い試験では低かったが，適度にしかそうでなかった． 我々は，警戒変動のデータ駆動推定値と作業記憶タスク性能との間の関連性を観察する．この観察は，警戒状態の変化から生じる人間の行動変動を予測する上で，先述の方法の潜在的な有用性を実証している．ワン・バック・パラダイムと短いISIのために，ここで使用されているタスク設計は，現在の比較に最適ではないことに注意する．例えば，現在の試行または前の試行のいずれかに先行する間隔内の低い警戒が，現在の試行での反応時間に影響を及ぼす可能性があるので，警戒レベルを照会する各試行の周りの時間枠は，いくぶん不確定である．現在，この研究をより敏感なタスクデザインで拡張している．

この発表で着目したのはワーキングメモリ課題において，fMRIとEEGを用いて解析を行っているところです．fMRIとEEGのα波とθ波比率の相関を求めて結果・考察を行っていて，reaction timesの速さとEVLの関係を見ていて興味深かったです．
 

発表タイトル    ：The neural substrate of the development of other race effect: An fNIRS study 著者                ： Guifei Zhou1, Jiangang Liu1 セッション名   ： poster session Abstruct   ： 人は自分の人種を他の人種よりも速く正確に認識します。このような人種・エフェクト（ORE）の原因の1つは、他の人種の顔よりも視覚処理の経験が豊富であることである。 OREは幼児期のように早期に観察され、小児期全体にわたって年齢の増加とともに発達する。しかし、ORE発症の実際の神経メカニズムはまだ不明である。 　浙江師範大学倫理委員会の承認を受けた今回の研究では、健康な右利き中国人の子供（61人の男性; 7.77±2.80歳）を実験前に両親または法的保護者の書面による同意を得て募集した。実験には中国語の顔（ChF）タスクと白人顔（CaF）タスクが含まれ、学習フェーズとテストフェーズが含まれている。学習段階では、参加者に10人のターゲット面を覚えてもらうように求めた。次に、テスト段階では、これらの10個の「古い」面が、追加の10個の「新しい」面と混合されました。参加者は、これらの顔が学習段階で見られたかどうかに応じる必要がある。 2つのタスクの順序は、参加者間で釣り合っていた。機能的な近赤外分光法（fNIRS、46チャンネル、ETG-4000、日立メディカル）によって取得された機能的データは、NIRS_SPM を用いて前処理された。次に、すべてのチャネル対のfNIRS時間経過の相関係数を計算することによって、ChFおよびChFタスクのために機能的接続ネットワーク（FCN）をそれぞれ構築した。神経OREは、ChFタスクのそれからCaFタスクのFCNを引くことによって定義された。次いで、神経OREの相関係数および参加者の年齢を算出した。ブートストラッピングベースの方法を用いて相関強度の信頼性を保証し、誤検出の可能性を低減した[7]。まず、無作為に選択された50％の参加者に基づいて、年齢と神経OREとの相関を計算した。そして、このステップを10000回繰り返すことにより、平均相関強度を算出した10,000個の相関係数が得られた。順列試験を行った。まず、参加者の50％をランダムに選び、参加者の年齢をランダムにスクランブルした。次に、上記と同様の相関分析を行い、スクランブルデータの相関強度を求めた。このステップを10,000回繰り返すことによって、10,000の相関係数が得られ、次いで降順でソートされた。スクランブルされていないデータの相関強度がこのオーダーの上位5％にある場合、それは有意であるとみなすことができる。 　6つの接続は、神経のOREと年齢との間に有意な正相関を示した（Ps <0.05）（図1）。これらの6つの接続のうち3つは、左のSFGdor（BA10）から右のMFG（BA45）まで、および左のIFGtriang（BA45）から、左のSFGdor（BA10）から右のMFG ）を右のSFGdor（BA9）に割り当てる。右側のIFGtriang（BA45）から右MFG（BA9）、右IFGoperc（BA6）から右MFG（BA9）、および右IFGoperc（BA6）から右への3つの接続は、右前頭回内にあったPreCG（BA6）。 　本研究では、子供の年齢が高ければ高いほど、中国人の顔よりも前頭葉の領域間の相互作用が多いことが分かった。人は自分の顔を自動的に認識しますが、人種レベルで他の人種の顔を分類する。したがって、自己の人種の処理は、他の人種の処理よりも多くの認知処理を必要とする。したがって、このような余分な処理をサポートするためには、他の人種の顔よりも自分の人種の顔を認識するために、より多くの神経資源が必要である。我々の知見はこの仮説と一致している。

この発表で着目したのは，fNIRSを用いて人種によって認知するスピードが異なるところです．年齢と共に視覚処理の経験が増えるため差が増大するそう．また結果の見せ方が上手だったので参考にしたいです．
 

発表タイトル      ：Brain-to-brain synchrony early in life: What can we learn from different hyperscanning techniques? 著者                ： Laura Astolfi, Hiroki Tanabe, Yasuyo Minagawa, Vanessa Reindl, セッション名      ： Symposia Abstruct       ：ハイパースキャン技術は、異なる対象の脳活動の同時記録を可能にする。過去数十年の間に洗練された新しいツールやテクニックが登場した今、脳内相関を研究することが可能になった。ユニークなシステムとして相互作用する被験者のグループの脳活動の間にある。ハイパースキャンは潜在的に画期的な新しいアプローチであり、典型的な社会的相互作用の変遷と発展を理解するための新たな視点を開拓しています。これらの新しい機会を考えると、未来の質問、異なるハイパースキャニング技術の現在の課題と限界を反映し、議論することは時機を得て重要であると思われる。これらには（1）異なる年齢群（乳児期から成人期）および神経画像技術（EEG、NIRS、fMRI）にまたがるハイパースキャンに適した実験課題のレビュー。 （2）方法論的アプローチ（周波数ベースの接続性（fMRIおよびNIRSで得られた血行力学的データに使用される時間相関およびグレンジャー因果関係の計算）、（3）被験者の特性（例えば、年齢および性別）の影響、神経同期測定について; （4）脳と脳との同期性の行動相関。このシンポジウムは、これらの問題やその他の問題の議論を刺激するフォーラムを提供する予定です。特に、生涯にわたる精神的健康のための初期社会的相互作用の関連性に関して、臨床的含意が強調される。要するに、このシンポジウムは、人間の発達過程における社会的相互作用のハイパースキャニング技術に関する最新の知識を提供することを目的としている。

この発表で着目したのは母子でHRFの立ち上がりが異なることです．母子だけでなく，個人によってHRFの立ち上がりが異なることが考えられるので，HRFの最適化も行っていく必要がると感じました．また，fNIRSの前処理ではウェーブレット解析を用いていた方が多く，今後の参考にしていきたいと感じました．解析手法についてももっと詳しく知りたいです．
学会参加報告書

報告者氏名	片山 朋香
発表論文タイトル
発表論文英タイトル	Functional connectivity analysis during breath-counting meditation using multichannel fNIRS
著者	片山 朋香，日和 悟，廣安 知之
主催	Organization for Human Brain Mapping
講演会名	The 23rd Annual Meeting of the Organization for Human Brain Mapping
会場	Vancouver Conventional Centre, Vancouver
開催日程	2017/6/25 ~　2017/6/29

 
 

1. 講演会の詳細

2017/06/25から2017/06/29にかけて，カナダのバンクーバー（Vancouver Conventional Centre,）にて開催されましたThe 23rd Annual Meeting of the Organization for Human Brain Mappingに参加いたしました．この学会は，ヒト脳の高次機能を様々なイメージング装置によって解明するために，最新かつ革新的な研究の情報を交換することや研究成果について議論することを目的に毎年開催されています．
私は25日から29日の全日に参加いたしました．本研究室からは他に廣安先生，日和先生，M2の萩原さん，石原さん，吉武さん，玉城さん，和田さん，M１の藤井さん，相本さん，三好さん，池田さん，石田翔也さん，中村圭佑さんが参加しました．
 

2. 研究発表
   • 発表概要

私は26日の午後のPoster Sessionに参加いたしました．発表の形式はポスター発表で，2時間自由に参加者の方と議論を行いました．
今回の発表は，「Functional connectivity analysis during breath-counting meditation using multichannel fNIRS」　という題目で，発表を行いました．以下に抄録を記載致します．

[Introduction] Mindfulness that directs our attention to the present moment without judgment is gaining popularity. Mindfulness meditation is expected cause improvements in our concentration and the skills of keeping a distance from negative things. It has consistently been reported that an expert’s brain state changes through meditation [1].In this study, we examined the functional connectivity among the brain regions of novices during the meditation state using functional near-infrared spectroscopy (fNIRS), which can be used in everyday life. [Methods] In this experiment, breath-counting meditation, which was easy for novices to practice, was employed. Breath counting involves concentrating on the consciousness by counting breathing. Ten meditation novice males participated in the experiment. Using 116-channel fNIRS, we measured the change in cerebral blood flow of novices during the resting and meditation states with their eyes closed. The measurement data were processed using a band-pass filter of 0.008–0.09 Hz, and all channels were associated with the brain regions parcellated using automated anatomical labeling. Correlation coefficient matrix, showing the functional connectivity between all brain regions during the resting and meditation states, was calculated. Fisher’s z-transformation was applied to the matrix to normalize the distribution of the correlation coefficients, and the functional connectivity during the resting and meditation states were compared using the degree corresponding to an edge density of 15%. [Results] The degrees of the right middle occipital gyrus (MOG), right inferior occipital gyrus (IOG), and right angular gyrus (ANG) were high during both resting and meditation states. The degree of the left calcarine sulcus (CAL) during the resting state and that of the orbital part of the left inferior frontal gyrus (IFG) during the meditation state were also high. As the MOG, IOG, and CAL are important in visual processing, it is presumed that the connections among these parts do not depend on the meditation state. As the ANG is related to the default mode network [2], novice subjects seem to have been mind-wandering during the resting and meditation states. On the other hand, since the degree of the IFG, which is important for response inhibition and attention, was high, subjects also attempted to inhibit mind wandering and keep attention to the task [3] [4]. Furthermore, the degrees of the right middle frontal gyrus (MFG) and right lingual gyrus (LNG) were significantly lower during the meditation state than during the resting state. The right MFG did not connect with the region near the CAL and LNG during breath counting. The LNG is related to visual processing. It is reported that the MFG activates during the meditation state and controls attention [5]. Because the regions relating to meditation activated more cooperatively during the meditation state than during the resting state, connections between these regions and the regions not related to the meditation state were reduced. [Conclusions] In this study, functional connectivity in novices during the resting and meditation states was examined. Breath counting was used as a meditation task, and the change in cerebral blood flow was measured using fNIRS. In many subjects, the connections between the right MFG and the regions related to vision processing were lower during the meditation state than during the resting state. In conclusion, these results suggested that functional connectivity among the regions unrelated to the meditation state was decreased through the meditation.

 

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
 
・質問内容1
東京医科歯科大学の方から視覚野が瞑想テンプレートに出ているが，瞑想に関係しているのかという質問を受けました．この質問に対して，視覚野は瞑想に関係していないが，閉眼でも活動すること，注意と関わりがある前頭部との結合は見られないため，独立的に活動していることが考えられると回答しました．今後，さらなる調査が必要だと考えています．
 
・質問内容2
質問者の氏名を控え損ねてしまいました．瞑想中の脳状態を定量化し，フィードバックすることを目的に行っているが，フィードバックはどのように行うのか，という質問を受けました．この質問に対して，まだ検討中であり，注意できている状態，マインドワンダリングな状態なのか判断し，自覚できるようにしたいと答えました．
 
・質問内容3
台湾の中央研究院統計科学研究所 郭柏志さんから脳の領域分割はどうやって行っているのかと質問をうけました．AALという脳の領域分割方法を用いて，被験者に空間的レジストレーションを行い対応した部位を割り出したと回答しました．今後，この分割方法は見直す必要があると考えています．
 
・質問内容4
同じく郭柏志さんから含有率の時系列データは一人の結果なのか，他の人も同様の結果なのかという指摘を受けました．これに対して，同様の結果になったと答えました．しかし，初心者や瞑想者の中にも違いがあると考えているため，今後，検討していきたいと考えています．
 
・質問内容5
質問者の氏名を控え損ねてしまいました．初心者はレストで含有率が高くなっているがどういうことなのか，という質問を受けました．初心者は瞑想できていないから，熟練者が瞑想時の結合が見られると回答しました．
 
・質問内容6
質問者の氏名を控え損ねてしまいました．含有率の結果に対して，統計検定をしていないのかという質問を受けました．N数が少ないため，検定はしていないと回答しました．今後，瞑想熟練者のデータを増やすのかという質問を複数の方からも受けたため，さらにNを増やす必要があると考えています．
 
・質問内容7
質問者の氏名を控え損ねてしまいました．閾値はなぜエッジ密度15％なのかという質問を受けました．英語で適切に回答することはできませんでした．今後，複数の閾値を試したいと考えています．
 
・質問内容8
東北大学の池田さんから，3人の瞑想テンプレートがばらばらで共通していないのはなぜなのかという質問を受けました．瞑想熟練者であっても瞑想方法が異なることが考えられると回答しました．今後，瞑想テンプレートの定義方法の再考，熟練者のNの増加が必要だと感じました．
 
 

• 感想

2回目の国際学会での発表となり，昨年と比べて落ち着いて準備や発表ができたと思います．今年は，NIRSのセッションではなく，瞑想などの認識や注意がテーマのセッションで発表しました．たくさんの方がポスターに興味を持ってくださり，昨年と比較してたくさん質問をもらうことができたと感じています．その中でも，共通してされた質問がいくつかあり，今後の検討の課題だと実感しました．
学会に参加することで，最前線の研究内容に触れることができ，自身の研究に足らないところ，今後どういうことをしていきたいか明確になったと思います．他の人のポスター発表を聞くことで，初めて聞く人にわかるように話す必要があることと，自分の発表にはそれが足りていないことがわかりました．質問への受け答えなど，英語力が足らないことを実感しました．しかし，昨年と比べてたくさん話すことができ，成長を実感できてよかったです．とても刺激を受けたので，次は修士論文に向けて研究を進めていきたいと思います．
 
 
 

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました．
 

発表タイトル       ：　Decoding Conversational Compatibility from Inter-Subject Correlation of Resting-State Networks 著者                  ： Shigeyuki Ikeda, Hyeonjeong Jeong, Yukako Sasaki, Kohei Sakaki, Shohei Yamazaki, Takayuki Nozawa, Ryuta Kawashima セッション名       ： Poster Session Abstract            ： Introduction:When we talk with a stranger, the conversation sometimes goes well, and sometimes goes wrong. Conversation success/failure may depend on compatibility of each other’s personality traits, cognitive states, and emotional states before the conversation. Resting-state brain activity is associated with the personality traits [Adelstein et al., 2011; Sugiura et al., 2000], and spontaneously represents cognitive states [Van Calster et al., 2016] and emotional states [Kragel et al., 2016]. The present study hypothesized that satisfaction level of conversation could be decoded from resting-state activities of two brains before dyadic conversation. To decode the satisfaction level, we applied multi-variate pattern analysis (MVPA) to the resting-state activities. Methods:Our study included 58 healthy, right-handed university students (age range: 18–23; 28 females). We obtained written informed consents from all subjects for their participation in this study. The Ethics Committee of Tohoku University Graduate School of Medicine approved this study. The subjects were randomly assigned pairs (29 pairs; 15 male–male pairs, 14 female-female pairs); members of each pair were unacquainted with each other before the experiment. In resting-state fMRI, a total of 180 functional volumes were acquired while each subject was resting. The subjects were instructed to keep still with their eyes closed, not to sleep, and not to think about anything in particular. After the resting state scan, the subjects completed the following sessions: a free-conversation session (3 minutes); three topic-conversation sessions (5 minutes for each session). In the free-conversation session, the pairs freely talked with each other to get comfortable with conversation. In each topic-conversation session, the pairs talked with each other based on a topic randomly chosen from three common topics (i.e., travel, hobbies, and school life). The pairs completed a questionnaire to assess satisfaction scores of the conversation (18 items; 8-point scale; Bernieri et al., 1996; Kimura et al., 2012) after each topic-conversation session. The satisfaction scores were averaged for each pair. The fMRI image preprocessing was carried out using SPM12. Linear trend, mean time courses (white matter and cerebrospinal fluid), friston 24 motion parameters were regressed out in 1st level analysis. Whole voxels were assigned to anatomical areas using Anatomical Automatic Labeling (AAL; 94 areas without cerebellar regions) [Rolls et al., 2015]. To obtain resting-state functional connectivity, mean time courses were calculated from voxels within each of the AAL areas; linear correlations between the AAL areas were calculated using the mean time courses. Inter-subject correlation of functional connectivity patterns (FCC; Conroy et al., 2013) was used as input features of MVPA. FCC represents the similarity (linear correlation) of functional connectivity patterns for corresponding nodes across subjects. To decode the satisfaction scores, we applied a linear support vector regression [Chang and Lin, 2011] to a FCC matrix (29 pairs × 94 nodes). To estimate generalization accuracy of the decoder, we used leave-one-pair-out cross-validation. Obtained results were assessed using permutation testing (10000 permutations). Results:The satisfaction scores of the topic-conversation sessions were decoded from FCC. The obtained results were validated by calculating linear correlations (r) and root mean squared error (RMSE) between decoded scores and the actual scores. We observed a significant result in the 1st topic-conversation session (r = 0.53, p = 0.01; RMSE = 0.46, p = 0.01). No significant results were observed in the other sessions. Conclusions:Our results showed that the satisfaction level of conversation between two strangers could be decoded from FCC measured before the conversation. FCC may, therefore, be a quantitative measure of conversational compatibility between two strangers.

初対面の人との会話の満足度はResting Stateの脳状態から解読できることを示唆した研究でした．Resting Stateは個人の特性によって活動の仕方が異なることはこれまでに論文を読んだことがあり，大変興味深かったです．シードベースでネットワークを個人ごとに算出し，個人のパターンを作成しどれくらい他の人と類似しているか見ている点は，私自身の研究にも参考になりました．現在，初心者，熟練者とグループに分けて検討しているので，それぞれの類似度を見るべきだと実感しました．この発表では，同性同士の関係をみており，今後異性同士などのペアを変えて関連を見るとのことでした．また，3つの会話で検討されていたのですが，被験者が複数になるとそういった実験設計や条件が増えるので難しいと感じました．
 

発表タイトル       ：　Developmental of Functional Brain Networks in the Early Children and Adolescents 著者                  ： Lin Cai, Haijing Niu セッション名       ： Poster Session Abstract            ： Introduction:Early childhood (7-8 years old) is a critical period when physical, social, and cognitive capacities of children develop quickly because children start to go to school and have to learn how to adapt themselves to school life. Correspondingly, Early adolescence (11-12 years old) is typically defined as a transitional period which is characterized by changes in social interaction, and cognitive development from immature children to independent adults(Spear, 2000). Therefore, it’s very pivotal to investigate brain development at these two stages. The extant literature focusing on the development of brain networks revealed that there was a general topological organizing principle guiding the organization of specialized functional networks shifts from a local anatomical emphasis to a more distributed architecture(Supekar, 2009). For now, two studies reported local efficiency increased from early childhood to adulthood but global efficiency was changeless(Wu, 2013; Cao, 2014). However, very little is known about how developmental patterns for connectome topology change at these two dramatic developmental periods. Here, we used resting-state fNIRS(Niu, 2013) and graph theory to address developmental changes at both global network metrics and regional nodal centrality metrics in human brain. Methods:We collected 10-min rs-fNIRS data from 30 early children(range 7-8 years), 30 early adolescents(range 11-12 years) and adults(range 19-27 years) by applying TechEn CW6 system with 46 measurement channels to cover almost the whole head. For each subject, we constructed whole-brain functional networks by computing Pearson correlation coefficients between each pair of channels. The correlation matrix was then thresholded into a binary matrix. We further analyzed global network metrics and nodal metrics using graph-theoretical approaches. Results:Economic small-world organization The functional brain networks of these three age groups consistently showed a higher clustering coefficient, local efficiency and modularity but similar characteristic path length and global efficiency compared with the matched random networks, respectively. The small-worldness of these three age groups was larger than 1 over the sparsity threshold(0.05<s<0.2). The nodal results showed that hubs were predominately located in the prefrontal and parietal lobes, similar to finding of a recent study[2]. These hubs may relate to high level cognitive functions and play a central role in information integration. Effects of age on global network metrics To characterize the age effect on each global network topological property, we separately performed one-way ANOVA on these three age groups. We found adults have a higher normalized clustering coefficient, normalized characteristic path length, and local efficiency than both children and adolescents(p<0.05), but there was no significance(p>0.05) between children and adolescents. Modularity of adults was stronger than children(p<0.05). There was no significant age effect on small-worldness, and global efficiency(p>0.05) (Fig. 1). Effects of age on nodal metrics The age-related increases in the nodal degree were predominately found in the frontal cortex (channel 2, 3, and 6), which was related to the increasing cognitive capacity from childhood to adulthood (Fig. 2). For the nodal efficiency, adolescents had higher nodal efficiency in parietal cortex (channel 26, 29, 32) which were primarily related to motor and somatosensory ability. Conclusions:Our results revealed that the functional brain networks dynamically optimize the integration of multimodal information and the segregation of local, specialized processing from early childhood, adolescence to adulthood. Effects of age on nodal properties suggested that the frontal brain regions associated with the higher-order cognitive functions developed from early childhood, whereas the parietal, brain regions tended to become less important.

fNIRSを用いて大人と子供のネットワーク解析をした研究でした．昨年と比較して，今年のOHBMでは，ネットワーク解析がfNIRS研究で多く見られたと思います．10分間のResting Stateを計測し，グラフ理論解析よりアプローチしていました．結合が増えているハブ，減少しているハブや他の特徴量を検討しており，研究室でしている検討と同じようなことが国際的にされていると感じました．デオキシヘモグロビンの機能的コネクティビティも検討しており，オキシヘモグロビンだけ見ているだけでよいのか疑問に思いました．
 

発表タイトル       ：　Neural Synchronization in lovers 著者                  ： yuhang long, Xialu Bai, Lifen Zheng, Hui Zhao, Wenda Liu, Chunming Lu セッション名       ： Poster　Session Abstract            ： Introduction:Romantic relationship is one of the most important relationship types in human society, plenty of studies have revealed unique features of romantic love. A lot of behaviors can reflect the particularity of people who are in romantic relationships. The eyes and their highly expressive surrounding region can communicate complex mental states such as emotions, beliefs, and desires.(Frischen, Bayliss, & Tipper, 2007). In addition, verbal communication also plays a significant role in romantic relationships (Gottman & Notarius, 2000). As the interactive nature of romantic relationship, it’s necessary to investigate two brains at the same time. Here, we used fNIRS-based hyperscanning to examine interpersonal neural synchronization (INS) of lovers when they were gazing and having a naturalistic verbal communication. Methods: Participants:29 pairs of lovers(mean age: 23±2y) and 30 pairs of cross-sex friends (mean age: 22±2y) were recruited by ads. Tasks and procedures:For each pair, an initial resting-state session of 5 min served as a baseline. Two task sessions immediately followed the resting state session. The two tasks were as follows: (1) gaze, (2) discussion. fNIRS data acquisition:A group of customized optode probe sets was used. The probes were placed around the lateral fissure on both the right and left hemispheres to cover the inferior frontal cortex and the temporal-parietal junction. Two optode probe sets were used on each participant in each pair. Each optode probe set consisted 13 measurement channels. CH11 was placed just at T3 in accordance with the international 10-20system. CH25 was placed at T4. Imaging data analysis:Wavelet transform coherence (WTC) was used to assess the cross-correlation between two fNIRS time series generated by pairs of participants as a function of frequency and time. The wavelet coherence MatLab package was used (Grinsted, Moore, & Jevrejeva, 2004). According to previous studies(Cui et al., 2012; Jing Jiang et al., 2015; J. Jiang et al., 2012), the coherence value increases when there are interactions between persons, compared with that during the resting state. In this study, gaze task induced subjects’ psychological activities in a low frequency, so we calculated the average coherence value between 0.02 and 0.04 Hz. As for discussion task, the interaction frequency was higher than gaze task, and the average coherence value between 0.09 and 0.16 Hz was calculated. Finally, the coherence value was time-averaged. The averaged coherence value of the resting-state session was subtracted from that of the task session, and the difference was used as an index of the INS increase between two persons. For each channel, after converting the INS increase into a Fisher z-statistic, a one-sample t test was performed on it across the participant pairs. Results:During gaze task, for lover pairs, a higher synchronization was found at CH24, which roughly covered the right superior temporal cortex, than during the resting-state condition[t(28)=4.61,p<0.0001, false discovery rate(FDR) correction]. No INS increase was found for any channels of the friends. Group differences between lover and friend pairs were significant at CH24 ( t(57)=3.07,p=0.003). During discussion task, the results showed a significant increase of neural synchronization at CH8 ( t(28)=3.48, p=0.0017, FDR correction) and CH13(t(28)=3.89, p<0.001, FDR correction), which roughly covered the left temporo-parietal junction, between lovers but not cross-sex friends. Conclusions:The results showed a significant increase of neural synchronization in the right superior temporal cortex during gaze task and left temporo-parietal junction during discussion task for lover pairs but not friend pairs. Our findings demonstrated that interactions between lovers were associated with higher-level interpersonal neural synchronization. These results will provide important insight into the neural mechanism of romantic relationship.

恋人同士のインタラクティブな脳状態を，fNIRSを用いてハイパースキャニングした研究です．恋人同士と友達同士で3種類の会話した時の活動の変化を検討されていました．Wavelet transform coherence を用いており，文献調査したときにNIRS研究に多く用いられているため，どのような解析方法か理解する必要があると思いました．この発表を聞いて，ハイパースキャニングの脳活動を検討するためには人の組み合わせによって大きく結果が異なると感じました．また，自分の研究として気になっている脳部位とチャンネルの対応付けは，国際10－20法を基準として設置しているとのことで，まだまだNIRS計測の検討すべき点だと感じました．
 

発表タイトル       ：　An fMRI analysis of reproducible brain activity while listening to real-world sounds 著者                  ： Po-Chih Kuo, Yi-Li Tseng, Philip E. Cheng, Michelle Liou セッション名       ： Poster Session Abstract            ： Introduction:The analysis of fMRI time series during real-world experience is methodological challenging because of the human brain processing a variety of features simultaneously in multiple-uncontrolled and -dynamic stimuli (Bartels et al. 2004). Previous studies have used block designs to localize the activated brain regions. Those experimentally defined control-states hardly exist in the real world. The reproducibility analysis is an approach to investigating event-free spontaneous brain activity. Here we designed an experiment engaging long-term auditory stimulation reflecting a real world experience. The reproducibility between two runs of the sound stimulation with either eyes-closed or -open as calculated using the proposed intraclass correlation (ICC) statistic, which is directly applicable to pre-processed fMRI time series. Methods:Thirty subjects (15 females, averaged age: 22.50±3.462 years) were instructed to passively listen to real-world sound stimuli with 4-min eyes-closed followed by 4-min eyes-open according to the acoustical instructions. The sound stimuli were consisted of human voices, including crying, laughing, guffawing, baby prattle, sneezing, and crowd talking, as well as animal sounds from a rooster, sheep, dog, cow or bird, along with ambient sounds from a farm. Each type of sounds was presented in a random interval [16, 26]sec and in a random order but the human voices were always presented first followed by the animal sounds. MR scan was performed using a 3T MAGNETOM Skyra scanner and a standard 20-channel coil. The EPI parameters whole-head coverage were: TR=2000ms, TE=30ms, slice thickness=3.4mm, FOV=192mm, voxel size=3×3×3.74mm, and T1 anatomical imaging parameters were: TR=2530ms, TE=3.30ms, FOV=256mm, voxel size=1×1×1mm. The fMRI time series were preprocessed in SPM8 and co-registered to the MNI space. The trends caused by the magnetic field drifts were also corrected. The fMRI time series were divided into two replicates, separately with eye-closed and -open conditions. The voxel-wise ICCs were computed and their standard deviations were also estimated for each individual subject. Given a voxel, the within-subject ICCs across the 30 subjects were finally synthesized into a Z value using the meta-analysis method. The phase-randomized procedure and FDR control with 5% family-wise Type-I error rate were used to find the supra-threshold voxels. The cytoarchitectonic parcellation provided by the ANATOMY toolbox (Eickhoff et al. 2005) was used to define active regions. Regions were clustered by using the hierarchical cluster analysis, in which the similarity between regions was evaluated by the correlation between brain activation patterns with the average linkage. Results:Figure 1 plots six brain regions with larger proportions of supra-threshold voxels compared with other regions (left Area s32: 96.9%, left Area TE 3: 96.9, left Area Fo1: 95.9, right Area OP1: 89.7%, right Area PFcm: 89.0%, and left Area TE 1.0: 86.7%). The fMRI time series in these regions were reproducible between eyes-closed and –open conditions, but brain responses to human and animal sounds (HS and AS) differ from each other. Figure 2 depicts the clusters of cytoarchitectonic regions in the left hemisphere along with clusters of sounds induced similar responses in the brain. The right hemisphere provides similar results as those in Fig. 2. Conclusions:This study reports an ICC method to evaluate the reproducible brain activity while listening real-world sounds. The left anterior cingulate, superior temporal cortex, and orbitofrontal cortex are highly reproducible between the eyes-close and –open conditions compared with other regions. Moreover, induced responses to human voices and animal sounds are also distinct in these regions. The hierarchical cluster analysis further demonstrates a consistency between structural and functional brain parcellations, supporting the cytoarchitectonics utility in probing the functional brain.

fMRIを用いて人の声と動物の声を聞かせた時の，閉眼，開眼実行時の再現性の検討をしたものでした．自身の研究で閉眼時に視覚野が活動しているので，閉眼，開眼でどのように変化するか興味ありました．この発表で，脳の領域間の結合の類似性と活性化パターンの相関をクラスタ分析した結果，人と動物の音声に対する脳の反応は異なったことが図より示唆されていることが興味深かったです．再現性に着目して脳活動を評価しており，自身の研究においても調査したいと思いました．
 

発表タイトル       ：　Brain Network of Emotion Regulation in Soldiers with Trauma 著者                  ： D Rangaprakash, Michael Dretsch, Thomas Daniel, Thomas Denney, Jeffrey Katz, Gopikrishna Deshpande セッション名       ： Emotion and Motivation Abstract            ： Introduction:Our ability to shape our emotional experience is termed emotion regulation (ER) [1], involving voluntary modification of emotions elicited in response to exogenous stimuli. Several functional MRI activation studies have consistently identified the middle frontal gyrus (MFG), anterior cingulate and insula to be involved in it [1]. Their limitation lies in the inability to explain the interrelationship between these regions, i.e. connectivity. The brain network of ER either in healthy adults or in psychiatric disorders like posttraumatic stress disorder (PTSD) and mild-traumatic brain injury (mTBI) has been elusive. Emotion dysregulation (ED) is regarded a primary cause for many symptoms observed in PTSD and mTBI [2]. Using fMRI data collected during an ER task, we obtained the network of ER in healthy soldiers and ED in soldiers with comorbid PTSD and post-concussion syndrome (PCS, or chronic mTBI). Methods:59 male U.S. Army soldiers were recruited (comorbid PCS+PTSD=36, combat controls=37, matched in age, race and education). FMRI data was acquired in a Siemens Verio 3T scanner (EPI sequence, TR/TE=600/30ms, flip-angle=55o, voxel size=3.5×3.5×5mm3). The ER task (Fig.1) was similar to Urry et.al. [3]. Participants were presented images eliciting a negative emotional response, and were asked to either “maintain” their emotional response, or “suppress” it (reduce negative feelings, requiring ER). There were 4 task blocks, with 24 trials in each block. Standard pre-processing was performed in SPM (realignment, smoothing [8mm kernel], normalization to MNI space). We first identified significantly activated regions during ER (see Fig.2 for region selection procedure). Hemodynamic deconvolution was performed [4] on mean time series extracted from identified regions, to minimize the non-neural intra-subject HRF variability [5]. We employed effective connectivity (EC) modeling using Granger causality (GC) [6] to assess directional causal relationships between identified regions, similar to recent works [7]. Subject-wise EC between all regions were obtained, using which the networks of ER in healthy soldiers (suppress>maintain) and its impairment in PCS+PTSD (control>PCS+PTSD for ‘suppress’ condition) were obtained (p<0.001, Bonferroni corrected) (Fig.3). We provide novel evidence for the brain networks of both ER and ED in a clinical population. Results:We investigated brain networks of ER in healthy soldiers, and ED in PCS+PTSD. We defined our ROIs around the 9 regions activated during the ER task (Fig.4). With EC analysis, we found the ER network having a top-down structure with the MFG driving the rest of the network (insula, medial prefrontal, amygdala and lateral parietal regions) (Figs 5a,5b,5c). During ED this network was imbalanced with reduced prefrontal connectivity and elevated subcortical and lateral parietal connectivity (Figs 5d,5e,5f). Our ER network fits well with prior findings [1, 8], which identified the pivotal role of MFG in the initiation of ER. MFG is implicated in executive functions like cognitive control [1], which are necessary for regulating emotions. Soldiers with PTSD exhibit impaired emotional processing [9] and impaired cognitive functions associated with the MFG [10],. All directional connections are traceable to the MFG, implying that MFG could be the source of ER [1]. As for ED, the MFG emerged as the key source of disruption in PCS+PTSD. All connections from MFG had reduced connectivity, whose “ripple-effect” culminated in disinhibition of amygdala, which might translate to symptoms like flashbacks, trauma re-experiencing and hyperarousal. This fits well with behavioral manifestations of these conditions [2]. Conclusions:In summary, we identified the MFG as pivotal to ER in healthy soldiers and ED in PCS+PTSD. Our findings are significant given that these regions are implicated in prior activation studies [1, 8], but a precise understanding of the underlying network structure and their causal relationships had not emerged so far.

 
fMRIを用いて，軍人にニュートラルな画像と不快な画像を見せて感情調節のネットワークを調査した研究でした．グループ，個人，メタ分析のそれぞれとどのように活動が一致しているのか，感情制御ネットワークと制御できていないネットワークがどのようになっているかなどの話が大変興味深かったです．グレンジャーの因果性を用いて領域感の方向性の因果関係を評価しており，ただどのような領域が結合しているかだけではなく，因果関係も調査する必要があると思いました．MFGを起点として，中心的な役割をしているネットワークだとわかりました．
 
 
参考文献

• OHBM 2017 Annual Meeting,

https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageID=3734
 
学会参加報告書

報告者氏名	相本武瑠
発表論文タイトル	学会参加報告書
発表論文英タイトル	Conference Report
著者	相本武瑠,日和悟,廣安知之
主催	医療情報システム研究室
講演会名	Organization for Human Brain Mapping
会場	Vancouver Convention Centre
開催日程	2017/06/25-2017/06/29

 
 

1. 講演会の詳細

2017/06/24から2017/06/29にかけて，Vancouver Convention Centreにて開催されましたOHBMに参加いたしました．このOHBMは，ニューロイメージングを用いて人間の脳の組織を解明しようとする様々な国際組織によって主催された研究会で，人間の健康的な、もしくは病変を有する脳の解剖学的，機能的組織の理解を発展させることを目的としています．
私は全日参加いたしました．本研究室からは他に廣安先生，日和先生，石原さん，玉城さん，和田さん，吉武さん，片山さん，萩原さん，石田（翔），三好，中村（圭），池田，藤井，水野が参加しました．
 

2. 研究発表
   • 発表概要

私は27日の12:45~14:45のセッション「Poster Session : Poster #’s 1000-2223」に参加いたしました．発表の形式はポスター発表となっており，2時間にわたり，参加者の方と議論を行いました．
今回の発表は，「Intra-individual variations in functional connectivity during resting and meditative states」と題して発表いたしました．以下に抄録を記載致します．

Introduction In recent years, studies measuring the brain states of expert meditators have been conducted using fMRI. It is important to grasp how much each brain state has the variations when examining changes in the brain state through the meditation and the difference between individuals. In particular, beginner meditators are expected to have large brain state variations. Here brain activities during resting and meditative states were repeatedly measured using fMRI, and intra-individual variations in functional connectivity were examined using graph-theoretical metrics. Methods This experiment used breath-counting meditation, in which the practitioner focuses on counting their own breath, because beginners can easily perform this type of meditation. Seventeen healthy adult beginner meditators were assessed, and two of them (subjects A and B) were assessed 10 times on a separate day. Whole brain scans were parcellated into 116 regions using automated anatomical labeling (AAL). ROI-wise functional connectivity was calculated for resting and meditative states. Betweenness, degree, and eigenvector centralities were computed using graph theoretical analysis. The unbiased variance of each graph theoretical metric was calculated for both the group of 17 subjects and for subjects A and B, and these values were compared. Results A test for equality of variance was performed between the individual subjects A or B and the group data. The brain regions where intra-individual variation was significantly larger than the group are shown in Table 1. Betweenness centrality of Cingulum_Post_L, degree centrality of Cerebellum_8_R, and eigenvector centrality of Caudata_R were shown to have large intra-individual variances in both subjects A and B during the resting state. During the meditative state, degree and eigenvector centralities of Putamen_L, eigenvector centrality of Putamen_R, Pallidum_L, and Pallidum_R had large intra-individual variance. Putamen_L and R are related to the recall of negative memories. Pallidum_L and R are involved in motor control and motivation. Because these regions play important roles in meditation, these results suggest that beginners show intra-individual variability in the quality of meditation. Conclusion Here the intra-individual variations in resting and meditative brain states were investigated using graph theoretical metrics of functional connectivity. Brain regions with significantly larger intra-individual variation than group data were confirmed. These data suggest that variations within these brain regions must be considered when quantitatively evaluating the brain state of beginner meditators in a group analysis.

 

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
 
・質問内容1
質問は「Resting-stateにおいて前部帯状回のばらつきが大きいと述べている文献はあるのか．」というものでした．この質問に対して私は、「今現在のところ確認していない．」と回答しました．質問者の氏名は控え損ねてしまいました．
 
・質問内容2
質問は「なぜ20人中2人しか10回測定していないのか．」というものでした．この質問に対して私は、「この実験はまだ完了していないため．」と回答しました．質問者の氏名は控え損ねてしまいました．
 
・質問内容3
質問は「AALはどのようにして領域を分割しているのか．」というものでした．この質問に対して私は、「原理までは理解していない．」と回答しました．質問者の氏名は控え損ねてしまいました．
 
・質問内容4
質問は「なぜエッジ密度15%を使用しているのか．」というものでした．この質問に対して私は、「それが脳のスモールワールド性を保つ効率的な値のため．」と回答しました．質問者の氏名は控え損ねてしまいました．
 
・質問内容5
質問は「何と何の分散値を比較したのか．」というものでした．この質問に対して私は、「集団と個人の脳領域毎の特徴量の分散値を比較した．」と回答しました．質問者の氏名は控え損ねてしまいました．
 
・質問内容6
質問は「特徴量について簡単に教えて欲しい．」というものでした．この質問に対して私は、「どの特徴量も値が高い時に，その領域が脳機能において中心的な役割を持つ．」と回答しました．質問者の氏名は控え損ねてしまいました．
 

• 感想

初めての国際学会でとても緊張した．質問に来てくれた人達の質問内容はおおよそ聞き取ることが出来たが、それに応える語彙力や表現力が足りていないことがわかった．来年の国際学会に向けて、しっかりと研究を行い、また英語力も向上させたい．
 

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました．
 

発表タイトル        ：　Brain mechanisms underlying symptom improvement in chronic visceral pain after mindfulness training   著者                    ： Ravi Bhatt, Jennifer Labus, Cody Ashe-McNalley, Arpana Gupta, Suzanne Smith, John Serpa, Jean Stains, Bruce Naliboff, Kirsten Tillisch   セッション名    ： Poster session Abstract： Introduction: Background: Irritable Bowel Syndrome (IBS) is a brain-gut disorder characterized by abdominal pain that is associated with altered bowel habits. IBS patients have functional brain alterations in regions associated with salience and emotional processing. (1) Mind-body interventions, such as hypnosis, cognitive behavioral therapy and Mindfulness Based Stress Reduction (MBSR) have been used to successfully treat symptoms in IBS, though the mechanism of this improvement is not known. (2) Aims: Discover symptom-related changes in resting state network connectivity (RS-FC) in patients with IBS who have undergone a 9 session MBSR intervention. Methods: Methods: Men and women aged 18-55 were recruited by advertisement and from clinics at UCLA. A high resolution T1 structural image and 10-minute eyes closed resting state fMRI was performed on a 3T Siemens scanner (TE: 28 ms, TR: 2000 ms, flip angle: 77 degrees, FOV 220mm x 220 mm, acquisition matrix: 64 x 64, slice thickness 4.0mm with a 0.5mm skip) before and after the MBSR intervention. The intervention consisted of eight 2 hour visits and 1 half day retreat using a standardized MBSR model (3). Mindfulness was measured using the Mindful Attention Awareness Scale (MAAS), and IBS symptoms with the IBS-Severity Scoring System (IBSSSS). Structural images were segmented and parcelled into 165 regions based on Destrieux and Harvard-Oxford atlases. ROI-to-ROI FC analysis was performed in the CONN toolbox. The function network matrix was comprised of z transformed r scores thresholded at z>.3. Network analysis via graph theory was applied using in house MATLAB code and the GTG toolbox to compute the functional network centrality of emotional processing (amygdala) and salience (anterior insula [long gyrus, short gyrus, circular sulcus]) regions. Network centrality indices included Degree strength, Betweenness Centrality, and Eigenvector centrality. Post-Pre intervention change scores in IBS-SSS and network centrality indices were correlated and significance was considered p<.05 corrected using false discovery rate. Results: 63 subjects (47 females) completed MBSR training and both scans. Mean age was 33 y (SD=9.80 19-54 years). The mean improvement in IBS-SSS from first to second scan was 74.8 (t(61) = 5.57, p < .001), with a 50-point change being considered clinically significant. The MAAS increased by 2.5 (t(59) = 2.41, p = .02). Decreased network centrality of the amygdala and the anterior insula after MBSR was associated with IBS symptom improvement and increased mindfulness (See Figure 1). ·Decreased network centrality of the amygdala and the anterior insula after MBSR was associated with IBS symptom improvement and increased mindfulness. Conclusions: IBS patients undergoing an MBSR intervention have improvements in mindfulness and overall IBS symptoms. These improvements are associated with decreases in emotional processing and salience regions.

本発表は，MBSR訓練の前後の脳状態を，グラフ理論特徴量であるstrengthや，eigenvector centralityを用いて計測したものでした．脳の状態を考察する上でこれらの特徴量も検討すべきであると感じました．

発表タイトル       ：Longitudinal evaluation of military training stress effects on white matter diffusion metrics 著者                    ： Nicholas Davenport1, Kelvin Lim, Erin Begnel セッション名    ： Poster session Abstract              ： Introduction: Prolonged exposure to stressful circumstances is hypothesized to have deleterious effects on brain connectivity. Military training provides a unique opportunity to test this hypothesis with a prospective longitudinal investigations of a set of experiences that are consistent in intensity, nature, and duration across individuals. Moreover, the limited age range in which individuals undertake training provides an additional opportunity to determine how these effects interact with normal brain development. Methods: As part of ongoing data collection for a large study, a wide range of MRI, clinical, and self-report measures have been collected from 135 newly enlisted Minnesota Army National Guard service members (mean age 20.8 years) within 1 month of shipping to 16-20 weeks of basic (i.e., “boot camp”) and advanced individual training (AIT), and follow-up data were collected from 35 of these individuals after returning. Additionally, upon returning from training, service members completed a mailed survey rating stress perceptions of training experiences. Measures of white matter diffusion properties, including anisotropy (FA), diffusion magnitude (MD, RD, AD), kurtosis (RK, MK), and complexity (ODI), were calculated based on multi-shell diffusion data. Longitudinal effects of training stress on white matter were investigated through correlations between changes in these measures and ratings of various sources of stress. Additional correlations between these measures and psychological scales of personality, psychopathology, and cognition were also explored in the large baseline data set. Results: Few interpretable relationships between changes in white matter diffusion and ratings of training stress were observed, suggesting that the magnitude of stress is insufficient to markedly affect brain circuitry. However, relationships between FA and age were observed cross-sectionally at baseline and longitudinally across the training time period, suggesting that diffusion imaging is sensitive to developmental changes within this limited time window (ages 18-22). Conclusions: Despite evidence that military training is perceived, by at least a subset of individuals, as stressful, and that the level of perceived stress is associated with increases in depression symptoms, these effects are not reflected in altered brain connectivity. Given that these diffusion MRI measures are sensitive to developmental changes within a limited age range, it is possible that stress-related effects are overshadowed by normal age-related changes.

本発表は，長期に渡る軍事訓練のストレスが，脳構造にどのような影響をもたらすのかを調査したものでした．結果として，脳の帯状回の体積が減少しており，ストレスと帯状回の関係をより検討していく必要性があると感じました．
 

発表タイトル    ：Successful encoding activation modulated by empathic traits in memory for highly empathetic people   著者                ：Natsumi Kondo, Hikaru Sugimoto, Takashi Tsukiura   セッション名    ： Poster session Abstract          ： Introduction: Empathic ability is crucial in understanding intentions of others. Previous studies have demonstrated that personality traits of empathy are correlated with individual abilities of memory (Beadle et al., 2013; Wagner et al., 2015). However, little is known about the neural mechanisms underlying the empathy-memory interaction. The present fMRI study investigated encoding success activation (ESA) modulated by multiple kinds of empathic trait in memory for highly empathetic people. Methods: Twenty-four right-handed, college-aged healthy women participated in this study (mean age: 21.8, SD: 1.6). All participants were recruited from the Kyoto University community, and paid for their participation. They gave informed consent to a protocol approved by IRB of the Graduate School of Human and Environmental Studies, Kyoto University. All participants performed both encoding and retrieval tasks, and neural activation was measured only in the encoding phase. During encoding, participants were presented with pairs of an unfamiliar face and a sentence describing hypothetical action, and were required to rate how empathetic the faces presented with the hypothetical actions are. After the encoding, participants were presented with previously learned and new faces one by one, and were required to recognize whether each face was learned in the encoding phase. In addition, individual traits of empathy were evaluated by the affective and cognitive empathy (Davis, 1980; Sakurai, 1988) and the Japanese version of EQ-SQ questionnaires (D-score) (Baron-Cohen et al., 2003; Baron-Cohen & Wheelwright, 2004; Wakabayashi et al., 2006). All encoding trials were divided into highly empathetic (High) and low empathetic (Low) faces by subjective ratings during encoding, and all High and Low trials were subdivided into subsequent hits (H) and misses (M). ESA was identified by H vs. M in each condition of High and Low, and the empathy-related enhancement of ESA for face memories was identified by comparing between ESA in the High and Low conditions. In addition, we investigated correlations between the empathy-related enhancement of ESA and each empathic trait of the affective and cognitive empathy, and the D-score. All MRI data were acquired by a Siemens MAGNETOM Verio 3T MRI scanner. A gradient echo EPI sequence for functional images was employed by the following parameters (TR=2 s, TE=25 ms, flip angle=70 degree, 39 slices, 3.5 mm slice thickness). The preprocessing and statistical analyses for all functional images were performed by SPM12. Results: In behavioral data, response time (RT) during both encoding and retrieval was significantly smaller in High than in Low (Encoding: F=6.40, p<.05, ηp2=.22; Retrieval: F=7.62, p<.05, ηp2=.25). In addition, RT during the successful retrieval of highly empathetic faces was significantly smaller than that in the other conditions (F=6.18, p<.05, ηp2=.21). fMRI data in the regression analyses demonstrated that the empathy-related enhancement of ESA in a posterior part of the left dorsomedial prefrontal cortex (dmPFC) was positively correlated with individual score of the affective empathy, and that the empathy-related enhancement of ESA in an anterior part of the left dmPFC was positively correlated with individual score of the cognitive empathy. In addition, a significant correlation between the empathy-related enhancement of ESA and individual D-score was identified in the right temporoparietal junction (TPJ). Conclusions: The present findings suggest that ESA increased in highly empathetic faces could be associated with three different regions of the posterior dmPFC, anterior dmPFC, and right TPJ, each of which reflected individual difference in the affective empathy, cognitive empathy, and EQ-SQ difference (D-score). The enhancing effect on responses for highly empathetic people in memory-related processes could be modulated by several different components of empathic traits.

本発表は，共感の記憶に関連する脳内の神経メカニズムを賦活解析と，脳機能解析を用いて解明していました．賦活解析と，脳機能解析を同時に行うことの必要性を，改めて感じました．

発表タイトル    ：Large-scale functional connectivity networks predict attention fluctuations 著者                ： Monica Rosenberg, Dustin Scheinost, Wei-Ting Hsu, Emily Finn, R Constable, Marvin Chun   セッション名    ： Poster session Abstract          ： Introduction: Attention is crucial to navigating almost every aspect of daily life. However, we lack a good way to measure people’s attentional abilities as a whole. To address this challenge, recent work demonstrated that a person’s unique pattern of functional brain connectivity can index the ability to sustain attention (Rosenberg et al., 2016a). In particular, a model based on large-scale functional connectivity networks − the sustained attention connectome-based predictive model − predicts how well people can focus in a variety of contexts, generalizes to predict abilities in novel people and groups, and predicts attentional changes resulting from pharmacological interventions (Rosenberg et al., 2016b). Methods: To investigate whether, in addition to predicting individual differences in attention, network models predict attention dynamics, we used models based on large-scale connectivity networks to predict fluctuations in attention within individuals. FMRI data were collected as 25 participants performed a sustained attention task, the gradual-onset continuous performance task (gradCPT; Esterman et al., 2013; Rosenberg et al., 2013). Participants performed the task during 3 functional imaging runs; each run consisted of 4 3-min gradCPT blocks. Performance during each block was assessed as d’ (sensitivity); overall performance was calculated as average d’ across all 12 blocks. Predictive models were defined using connectome-based predictive modeling (Shen et al., in press; Finn et al., 2015). Briefly, network nodes were defined using a 268-node brain atlas (Shen et al., 2013). Connectivity matrices were computed by correlating the average BOLD signal timecourse of every pair of nodes. For each participant, one overall connectivity matrix was calculated using all volumes acquired during task performance, and 12 block-specific connectivity matrices were computed using volumes acquired during each gradCPT block separately. To identify attention-relevant connections, robust regression was performed between each connection in the overall connectivity matrices and overall d’ scores across n–1 subjects (the training set). Connections positively and negatively correlated with task performance at p < .01 were retained for model building. Linear models were defined relating network strength − that is, the sum of the functional connections − in the positive (high-attention) and negative (low-attention) networks to overall d’ in the training set. These models were then used to predict each block-specific d’ score of the left-out individual. In other words, network strength was calculated using each of the left-out subject’s 12 block-specific connectivity matrices, and input into models defined using data from the other 24 subjects to generate a predicted d’ score for each block. If network models are sensitive to fluctuations in attention, predicted and observed d’ scores should be correlated within subject. Results: Demonstrating that network models are sensitive to local fluctuations in attentional performance, predicted and observed d’ scores were significantly correlated within subject (mean r-value = 0.47; t(24) = 5.72, p = 6.8e–6). Significance was determined by comparing within-subjects correlation coefficients to chance with a paired t-test. Chance was determined separately for each individual via permutation testing. Comparable results were observed when high- and low-attention network strength were calculated in real-time (i.e., during data collection) in an independent group of participants, suggesting that these networks could be targets for interventions such as real-time neurofeedback. Conclusions: The current results demonstrate that functional brain connectivity is a robust measure of fluctuations in attention during task performance, and thus a useful index of attentional abilities as a whole. Furthermore, results suggest that the same networks that vary with attentional abilities across people vary with changing attention function within single individuals.

本発表は，脳機能ネットワークを用いて個人内と個人間の注意状態を推定するものでした．個人内の変動を，タスクをいくつかのブロックの分け解析する方法が，自身の研究でも活かせるのではないかと考えました．

発表タイトル        ：Can brain state be manipulated to emphasize individual differences in functional connectivity? 著者                ： Emily Finn, Dustin Scheinost, Daniel Finn, Xilin Shen, Xenophon Papademetris, R Constable セッション名    ： Poster session Abstract          ： Introduction: While neuroimaging studies typically collapse data from many subjects, brain functional organization varies between individuals, and characterizing this variability is crucial for relating brain activity to behavioral phenotypes. Rest has become the default state for probing individual differences, chiefly because it is easy to acquire and a supposed neutral backdrop. However, the assumption that rest is the optimal condition for individual differences research is largely untested. In fact, other brain states may afford a better ratio of within- to between-subject variability, facilitating biomarker discovery. Here, we present analyses of within- and between-subject variability across a wide range of scan conditions to determine if and how brain state can be manipulated to emphasize individual differences in functional connectivity. Methods: Data were obtained from the Human Connectome Project (Van Essen et al., 2013), 900 subjects release. Analyses were limited to 716 subjects that had complete data for each of nine functional scans: EMOTION, GAMBLING, LANGUAGE, MOTOR, RELATIONAL, REST1, REST2, SOCIAL and WORKING MEMORY (WM). Using a 268-node functional brain atlas, for each subject, we calculated nine connectivity matrices consisting of the pairwise correlation coefficients between each possible pair of nodes using data from each scan condition, respectively. Because connectivity matrices are symmetric, we extracted the unique elements by taking the upper triangle of the matrix; this results in a 1×35,778 vector of edge values for each subject for each condition. These vectors can then be compared using Pearson correlation either between different subjects in the same condition (yielding a 716 x 716 between-subject correlation matrix for each condition, with 255,970 unique values representing similarity between all possible subject pairs), or within the same subject across conditions (yielding a single 9 x 9 within-subject correlation matrix for each subject). Results: Our analysis showed that brain state does affect between-subject variability. The RELATIONAL task had the highest between-subject similarity (r = 0.53), while the two REST sessions, along with the MOTOR session, had the lowest between-subject similarity (r = 0.35). Given equal scan durations, two different tasks sometimes showed higher within-subject similarity than the two rest scans. Mean similarity between the RELATIONAL condition and the EMOTION, GAMBLING and WM conditions (r = 0.62-0.64) all exceeded mean similarity between REST1 and REST2 (r = 0.55). We also replicated the identification experiments described in Finn et al. (2015), in which a target matrix from one scan condition was used to identify the same individual from a set of matrices from a different scan condition. While rates were well above chance for all condition pairs, some pairs were more successful than others (accuracy range = 15%-92%). Interestingly, conditions that made subjects look more similar to one another tended to make better databases for identification experiments (r = 0.82, p = 0.007). Conclusions: We present these observations as proof-of-principle that individual differences in functional connectivity do, in fact, depend on the condition in which they are measured. We hope these results provide a jumping-off point for more detailed investigations into how brain state affects both within- and between-subject variability, which will help determine which conditions are optimal for individual differences research. Depending on the trait or behavior under study, certain tasks may bring out meaningful idiosyncrasies across subjects, essentially enhancing the individual signal in networks of interest beyond what can be measured at rest.

本発表は，さまざまなタスクにおいて，集団の脳状態が類似するか否かを検討するものでした．瞑想だけでなく様々なタスクにおいて個人間，個人内の脳状態のばらつきを検討することは重要であると感じました．
参考文献

• OHBM 2017 – Organization for Human Brain Mapping

, https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageID=3734
学会参加報告書

報告者氏名	石原知憲
発表論文タイトル	Optimizing electrode placement and frequency bands in EEG based motor imagery BCIs.
発表論文英タイトル	同上
著者	石原知憲,　 日和悟,　廣安知之
主催	Organization Human Brain Mapping
講演会名	OHBM2017
会場	Vancouver Convention Centre
開催日程	2017/06/26～30

 
 

1. 講演会の詳細

The Organization for Human Brain Mapping (OHBM) is the primary international organization dedicated to using neuroimaging to discover the organization of the human brain. The organization was created in 1995 and has since evolved in response to the explosion in the field of human functional neuroimaging and its movement into the scientific mainstream. One of the primary functions of the organization is to provide an educational forum for the exchange of up-to-the-minute and groundbreaking research across modalities exploring Human Brain Mapping.  It does this through a growing membership and an annual conference, held in different locations throughout the world.

2. 研究発表
   • 発表概要

私は27日の午後のポスター発表に参加いたしました．発表時間が120分となっておりました．
以下に抄録を記載致します．

Brain–computer interface (BCI) technology enables the control of an external device through brain activity without any physical movement. However, to perform the BCI operation, it is necessary to arrange a large number of channels (CH) for measuring the EEG, which increases the restraint time for and burden on the subject. In addition, the optimum bandwidth of the bandpass filter used in preprocessing, when classifying motor imagery using EEG, varies among individuals. To address these problems, in the present study, we investigated the usefulness of frequency band selection and CH selection for classifying motor imagery.

 

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
・質問内容1
「 CHを選択するごとに脳波の計測を行っているのか？ 」
質問者の氏名を控え損ねてしまいました．質問には，「CH選択は16CHを使用して計測し終わった脳波を用いて行っています．16CHのデータのうちどのCHの組み合わせが運動想起識別に最適かを探索しています．」とお答えしました．
 
質問内容2
「連続値のGAとはどういったものか?どのようにチャンネルの選択を指定しているのか？ 」
質問者の氏名を控え損ねてしまいました．質問には，「連続値GAとは遺伝子の遺伝子型が0-1表現ではなく，連続値で指定されるもので，その連続値の最適な組み合わせを探索するアルゴリズムです．チャンネルの指定は1ビットで指定される値を最も近い整数（0 or 1）に丸め込み，選択チャンネルの指定を行いました．」とお答えしました.
 
質問内容3
「今回GAで探索しているが，探索している組み合わせは全部で何通りあるのか？ 」
質問者の氏名を控え損ねてしまいました．質問には，「今回はCH選択数が3CH以下になるような条件で探索を行い，周波数帯域の組み合わせは1～40Hz幅で探索を行ったため，探索したすべての組み合わせは（16C3＋16C2）＊40C2　通りになる．」とお答えしました.
 
質問内容4
「 SVMは全員分のデータを学習して行っているのか？被験者ごとで識別を行っているのか？ 」
質問者の氏名を控え損ねてしまいました．質問には，「SVMの学習は個人内の運動想起時の脳波データのみで行っています．そのため被験者によって識別器の学習内容が異なります．」とお答えしました.
 
質問内容5
「 SVMの特徴空間（軸）とCSPで抽出された特徴量の関係は？ 」
質問者の氏名を控え損ねてしまいました．質問には，「CSPは2クラスのデータの分散比を最大化する固有ベクトルを算出します．具体的にはクラスAにあたる脳波の電位の分散値を最大化させ，クラスBにあたる脳波の電位の分散値を最小化させるベクトルとクラスAの分散値を最小化させクラスBの分散値を最大化させるベクトルが算出されます．この2つのベクトルによって射影・変換された特徴量を各軸にとりSVMで識別を行いました．つまりSVMの特徴空間の軸はCSPで算出された2つの固有ベクトルで抽出された特徴量となります．」とお答えしました.
 
 

質問内容6
「 時系列からどのように特徴抽出を行っているのか？ 」
質問者の氏名を控え損ねてしまいました．質問には，「今回はわかりやすく2CHを用いた際の例を説明します．まず各CHを軸とする軸空間に脳波の時系列データをプロットします．次に計算に用いられるCH離散データの集合にPCAをかけることで第２主成分までを求めます．ここで求めた第1・第2主成分を用いて各クラスに該当する脳波データに白色化処理を行い各クラスの脳波のプロットを直交させます．直交したプロットにさらにPCAをかけることでクラスAの第1主成分がクラスBの第2主成分に該当し，これらのベクトルを用いることで分散比を最大化することができ，識別に必要な特徴量を抽出することができます．」とお答えしました
 
質問内容7
「 遺伝的アルゴリズムとは何か？ 」
質問者の氏名を控え損ねてしまいました．質問には，「生物の進化を模擬した最適化アルゴリズムです．遺伝子の表現の仕方によりあらゆる問題に適用可能です．」とお答えしました．
 
質問内容8
「 どういう人向けのシステムなのか？ 」
質問者の氏名を控え損ねてしまいました．質問には，「重篤な神経疾患患者や体の自由が利かない患者さんのためのシステムとして研究を進めています．」とお答えしました．
 
質問内容9
「 BCIとは何か？ 」
質問者の氏名を控え損ねてしまいました．質問には，「脳信号を読み取り，脳と機械のダイレクトな情報伝達を仲介する技術です．今回はBCIのなかでも運動想起型BCI（MI-BCI）に焦点を当てています．」とお答えしました．
 
質問内容10
「 GAで組み合わせを探すうえで被験者数が少ないのではないか？ 」
質問者の氏名を控え損ねてしまいました．質問には，「今回は10人で解析していますが探索には不十分だと感じています．今後の課題として持ち帰ります．」とお答えしました．
 
質問内容11
「 選択されたCHの考察が難しそうだがどう解釈する？ 」
質問者の氏名を控え損ねてしまいました．質問には，「今回は電極位置とそれに関連する脳機能についてしか考察できていないので，最適化の結果，算出された結果の評価方法も含め今後の課題としていきたいです」とお答えしました．

• 感想

今回の学会は私にとって2度目の国際学会への参加でした．Motor Behaviorのポスターセッションで120分間のポスター発表を行いました．当日の発表では様々な国籍の方に発表を聞きに来ていただき，自分の研究をたくさんの人に聞いていただくことができました．2度目の国際学会でなおかつポスター発表ということで今回は「積極的に海外の人とコミュニケーションをとる」ことを目的に学会に参加しました．自分の英語に自信があるわけではありませんでしたが，初日のレセプションパーティから積極的に海外の学生さんとコミュニケーションを取ることができました．発表当日も学内での先生リハーサルで練習した時以上に，自分の発表を英語で伝えられたと自負しています．本学会では脳機能に関する研究，特にfMRIを用いた研究が中心だったため遺伝的アルゴリズムやEEGの研究に関するご指摘はあまりいただくことができませんでしたが，限られた時間内で自分の発表を英語で理解してもらえるように発表する力をつける事ができました．発表の中で，海外からお越しの方に質問をいただきましたが，質問の中の英語の意味が理解できず,答えることができないことも多々あり，悔しい思いをしました．この思いを踏まえてこれから英語の学習により一層力を入れて励みたいと考えています．この学会では研究の内容についての発見よりも，研究の進め方や深め方，英語の学習について学ぶことがたくさんありました．次の発表に挑戦する際には，一人で質問に正しく答えられるだけでなく，現地にて一人で生活できるぐらいには英語が話せるようになりたいと考えています．国際学会に参加し，学会期間中以外の経験も含め，たくさんの発見があり，毎日が大変有意義な時間でした．また学会に参加できるように毎日の研究と同時に，英語のリスニングとスピーキングの学習についても精進していきたいと考えています.
 

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました．

発表タイトル       ：　Combined Action Observation and Motor Imagery Neurofeedback for modulation of brain activity 著者                  ：　Christopher Friesen, Timothy Bardouille, Heather Neyedli, Shaun Boe セッション名       ：　Motor Behavior Abstruct            ： Motor imagery (MI) and action observation have proven to be efficacious adjuncts to traditional physiotherapy for enhancing motor recovery following stroke (Kim, 2014;Liu et al., 2014). Recently, researchers have used a combined approach called imagined imitation (II), where an individual watches a motor task being performed, while simultaneously imagining they are performing the movement (Tsukazaki et al., 2012;Wright et al., 2014). While neurofeedback (NFB) has been used extensively with MI to improve patients’ ability to modulate sensorimotor activity and enhance motor recovery, the effectiveness of using NFB with II to modulate brain activity is unknown.

この発表は運動スキルの学習において，運動想起に加え運動観察を同時に行うことで効果があるのかを実験している研究でした．まず運動観察を含むニューロフィードバック訓練を行い，その後運動想起をした際に感覚運動野でのERD/ERS値がどのように変化するかをニューロフィードバックのみで訓練した場合と比較していました．結果として提案されていた手法が有用であるという結論でした．被験者を訓練していく上で，脳波そのものを扱っているのではなく運動特性や運動学習過程に着目しているところが新しい知見でした．同じ生体信号である脳波を用いた研究でしたが人の運動機能回復のための運動というカテゴリでの研究で，異なる解析手法，アプローチであったため大変興味深いものでした．
 

発表タイトル       ：　Neural Synchronization in lovers 著者                  ：　yuhang long, Xialu Bai, Lifen Zheng, Hui Zhao, Wenda Liu, Chunming Lu セッション名       ：　Imaging method : NIRS Abstruct            ：　Romantic relationship is one of the most important relationship types in human society, plenty of studies have revealed unique features of romantic love. A lot of behaviors can reflect the particularity of people who are in romantic relationships. The eyes and their highly expressive surrounding region can communicate complex mental states such as emotions, beliefs, and desires.(Frischen, Bayliss, & Tipper, 2007). In addition, verbal communication also plays a significant role in romantic relationships (Gottman & Notarius, 2000). As the interactive nature of romantic relationship, it’s necessary to investigate two brains at the same time. Here, we used fNIRS-based hyperscanning to examine interpersonal neural synchronization (INS) of lovers when they were gazing and having a naturalistic verbal communication.

この発表は恋人関係の二人の脳活動の同調をNIRSの同時計測にて調査している研究でした．恋人関係は人間社会における最も重要な関係の1つであり，多くの行動はそういった関係にある人々の特殊性を反映するといわれています．この研究では恋人が凝視して自然に口頭でコミュニケーションを取っているときの対人神経同期（INS）を調べていました．恋人同士が3種類の議論しているときの脳活動の同期を見ており，脳活動を評価する指標にWavelet transform coherenceを用いていました．結果として恋人同士のペアでは left temporo-parietal junctionで，そうではないペアに対して差が出たというものでした．扱っている題材が恋人関係の人の脳活動的同期を見ているもので，研究の対象そのものが大変興味深いものでした．複数人のハイパースキャンを行う研究ではよくWavelet transform coherenceが用いられており，その説明を英語では聞き取りきることができなかったため，今後の課題として理解を深めていきたいです．
 

発表タイトル   ：　Human ECoG reveals dissocialable calculations for perceptual decisions and confidence judgement 著者                  ：　Megan Peters, Thomas Thesen, Yoshiaki Ko, Brian Maniscalco, Chad Carlson, Matt Davidson, Werner Doyle, Ruben Kuzniecky, Orrin Devinsky, Eric Halgren, Hakwan Lau セッション名       ：　Higher Cognitive Functions Other Abstruct            ：In humans and other animals, confidence judgements in perceptual decisions typically reflect the probability of the relevant decision’s being correct. However, the mere correlation between confidence and accuracy does not necessarily mean that the computation of confidence is strictly optimal. Here we sought to evaluate the neural representations involved, and whether the brain fully exploits all available information in computing confidence, as an ideal observer would, or whether some heuristic or approximating algorithm may be employed instead. Although some previous behavioral reports suggest confidence selectively relies on the magnitude of evidence favoring the decision while suboptimally down-weighting evidence favoring alternative, unchosen possibilities, this view remains controversial, and direct neurobiological evidence has thus far been lacking.

この発表は人間の物事の決定・判断を定量化し，そのメカニズムを解明しようとする研究でした．
タスクでは家なのかヒトなのかを被験者に判断してもらうタスクを行い，その時のECoGを解析する研究でした．具体的には計測したECoG信号から判断の脳活動をデコードしていました．発表を聞いて英語がかなり早くすべてを聞き取れている自信はありませんが，脳波と同種のECoG信号にあらゆる処理を施し，考察を深めていました．解析結果の見せ方やインパクトのある図，また人の判断と脳活動を紐づけるという研究の切り口などたくさんのことが新しい知見でした．個人的に予稿やポスターをよく読みこみ，今後の解析の参考にしたいと思いました．
 

発表タイトル       ：　FMRI study of working memory training 著者                  ：　Wan Zhao, Zhifang Zhang, Qiumei Zhang, Jun Li セッション名       ：　Working Memory Abstruct            ： Working memory is a cognitive system with a limited capacity that is involved in transient holding, encoding and manipulation of information (1, 2). fMRI studies have demonstrated that frontoparietal network is one of the most important brain mechanism for working memory (4-7). Working memory capacity can be expanded through targeted training (3). However, the brain mechanism of the training effect was still unknown till now. The current study tried to explore the contribution of working memory training on the brain activation.

この発表はワーキングメモリ訓練によって脳の構造がどう変化するかを調査した研究でした．17人の被験者は1日30分のトレーニングを20日間行い，MRIで脳画像を撮像していました．行動データの結果から訓練によりすべての被験者でワーキングメモリが拡張されたと報告されていました．脳の撮像データからはright parietal lobeとleft frontal lobeで活動の有意な増加が見られたと報告されていました． 医療情報システム研究室ではワーキングメモリについて研究がされていますが，被験者の訓練という研究はされていなかったため新しい知見でした．訓練を行う上で17人の被験者に1日30分×20日間訓練を行っているとのことでしたが，自分の研究においてもn数は根気強く増やし続けていく必要があると感じました．
 

発表タイトル       ：　Dissociable cortical contributions to the encoding of time and space information in episodic memory 著者                  ：　Saeko Iwata , Hikaru Sugimoto , Takashi Tsukiura セッション名       ：　Learning and Memory Other Abstruct            ： Episodic memory is defined as memory for personally experienced events with contextual information of time and space. Previous studies have reported that the lateral prefrontal cortex is involved in the retrieval of the temporal context (Suzuki et al. 2002; St Jacques et al. 2008), and that the retrosplenial cortex and parahippocampal gyrus are important in the retrieval of spatial context (Ekstrom et al. 2011; Burgess et al. 2001; Vann et al. 2009). However, available evidence is scarce in functional neuroimaging studies investigating neural mechanisms underlying the processing of temporal and spatial context during episodic encoding. The present fMRI study tackled this issue.

この発表はエピソード記憶が記憶として脳に定着する際に脳はどのように活動しているのかを調査している研究でした．実験タスクとしては時間情報，空間情報を提示し記憶させる課題，対照群としてニュートラルな情報を用いた実験が行われていました．エピソード記憶の定着具合は主観評価から得られる行動データで評価されていました．結果として記憶に関連する海馬とエピソード記憶に関連する部位とのネットワークが見られたというものでした．この研究ではネットワークを見る際にまず脳の賦活を考慮し，そこがハブとなっているネットワークを探すという解析を行っていたのは新しい知見でした．また解析を進める際に一つ一つの結果から仮説を立てて，その仮説を確認するために最適な実験方法をよく考えて行われているような印象を受けました．自分の研究を進めるうえでも，見習いたいと感じました．
 
参考文献

• OHBM2017 プログラム
• OHBM2017 https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageID=3734

学会参加報告書

報告者氏名	石田　翔也
発表論文タイトル	fMRIを用いた快・不快情動の機能的コネクティビティ解析
発表論文英タイトル	Functional connectivity analysis of pleasant and unpleasant states using fMRI
著者	石田翔也，日和悟，蜂須賀　啓介，奥野　英一，廣安　知之
主催	OHBM
講演会名	OHBM2017
会場	バンクーバーコンベンションセンター
開催日程	2017/06/25-2017/06/29

 
 

1. 講演会の詳細

2017/06/25から2017/06/29にかけて，カナダ，バンクーバーのコンベンションセンターにて開催されましたOHBM2017に参加いたしました．OHBMは，脳のイメージング装置を用いた人間の脳神経の解明において重要な組織であり，神経活動をマッピングするモダリティの最新かつ革新的な研究の交換のための教育フォーラムである．
私は全日程参加致しました．本研究室からは他に廣安先生，日和先生，三好，相本，中村，和田，玉城，石原，池田，水野，藤井，萩原，吉武，片山が参加しました．

2. 研究発表
   • 発表概要

私は27日のポスターセッションに参加いたしました．2時間の間ポスター前で質問に答えました．今回の発表の抄録を以下に記載致します．

Introduction: Excessive emotions can cause mental illnesses, such as depression. Self-monitoring enables the appropriate control and evaluation of one’s emotions. The appropriate control and evaluation of one’s emotions can affect learning abilities. Emotional expression is influenced by deep parts of the brain, including the amygdala and hippocampus[1]. Therefore, brain states during pleasant and unpleasant emotions were assessed using fMRI. Methods: The emotional experiment using the images of the Nencki Affective Picture System (NAPS) was performed[2]. Blood flow changes in subjects during pleasant and unpleasant states were measured using fMRI. The brain was parcellated into 116 regions with Automated Anatomical Labeling (AAL), and the correlation coefficient matrix was calculated betwen the blood flow changes in each region. Network features of degree centrality, clustering coefficient, and betweenness centrality of each brain region were calculated from the correlation coefficient matrix using graph theory analysis. A linear discriminant analysis (LDA) was performed with each 116-dimensional network feature to derive a discrimination vector which separates between the pleasant and unpleasant brain states. Prior to the LDA, feature variables used for classification were selected using the variable increment method. Next, the selected variables were manually reduced one-by-one so that erroneous discrimination did not occur. Brain regions related to discrimination were examined based on the element values of discrimination vectors. Results: The 116 regions were reduced to 48 regions utilized for LDA using the variable increase method. Finally, the brain regions used for discrimination in degree, clustering coefficient, and betweenness centrality were reduced to 20, 15, and 15 regions, respectively. Figure 1 shows the standardized element values for each discriminant vector. Figure 2 demonstrates the transition from the unpleasant state to the pleasant state by projecting the selected brain activity data on the discriminant axis. Each axis shows the discrimination axis of degree, clustering coefficient, and betweenness centrality, respectively. The degree of the opercular part of inferior frontal gyrus (F3OP) highly correlated with the discriminant axis. Notably, the positive correlation of the F3OP (right) shows that degree increases when changing from an unpleasant to pleasant state. On the other hand, the F3OP (left) has a negative correlation, which shows that degree increases from a pleasant to unpleasant state. Therefore, the F3OP shows a hemispheric difference in degree depending on the emotion. The parahippocampal gyrus showed hemispheric changes in the clustering coefficient depending on the emotion. Furthermore, the hippocampus (left) has a negative correlation with betweenness centrality. Thus, when an unpleasant emotion developed, the betweenness centrality became larger than the pleasant emotion and more stored information was exchanged. Conclusions: Brain activity data for pleasant and unpleasant emotions during image presentation were measured by fMRI. Two brain states were discriminated using LDA on each of three graph theoretical values, degree centrality, clustering coefficient, and betweenness centrality. The element values of discriminant vector for each feature value were associated with brain regions, and the regions related to pleasant and unpleasant emotions were obtained.

 

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
・質問内容1
この研究に期待する仮説は立てていないのか？
心拍や呼吸はなぜ計測しないのか？
・質問内容2
クロスバリエーションはしていないのか？

• 感想

OHBM2017は初めての学会参加であり，初の国際学会ということで緊張した．
オーラルセッションではFCの解析が当たり前に行われており，この分野の最先端な内容につい
て身をもって知ることができたと感じている．また，英語の能力の低さに気づくことのできた学会と
なった．しかし5日間の参加で多くの関連用語を学ぶことができた．

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました．
 

発表タイトル       ：　3761　FMRI study of working memory training 著者                  ： Wan Zhao, Zhifang Zhang, Qiumei Zhang, Jun Li セッション名       ： poster Abstruct            ： Introduction: Working memory is a cognitive system with a limited capacity that is involved in transient holding, encoding and manipulation of information (1, 2). fMRI studies have demonstrated that frontoparietal network is one of the most important brain mechanism for working memory (4-7). Working memory capacity can be expanded through targeted training (3). However, the brain mechanism of the training effect was still unknown till now. The current study tried to explore the contribution of working memory training on the brain activation. Methods:　This study’s protocol was reviewed and approved by the Institutional Review Board of the Institute of Cognitive Neuroscience and Learning at Beijing Normal University. 17 healthy participants recruited from the Beijing Normal University. All subjects were Han Chinese and gave written informed consent for this study. Subjects were trained on about half an hour (80 trials) per day for 20 days (4 weeks with five days each week) on the spatial span task which was designed according to Westerberg et. al.’s report (8). Subjects would enter into higher level after 5 successive correct responses. fMRI scans were performed before and after training. fMRI task was similar as the training task however only consisted of two conditions: memory condition (five green squares presented sequentially and in random order at every trial) and baseline (five red squares presented sequentially however in the same order at every trial) (Figure 1). There were 144 trials in total. All imaging data were acquired in a Siemens Trio 3T scanner at the Brain Imaging Center of Beijing Normal University. All image preprocessing and analyses were conducted using FSL 5.0.7 software. Contrast images (memory-baseline) for each subject were produced by first-level analysis. Higher-level analyses were carried out using single-group paired T test to calculate the effect of training (pre-training vs. post-training) across the whole brain. We reported all of our results at voxelwise P <0.005. The method of Alphasim was used to correct multiple comparisons. Results:　In whole-brain analysis, comparing with pre-training, post-training showed significantly increased activation at the right parietal lobe (cluster size= 108 voxels, peak voxel MNI coordinate: x= 46, y= -34, z= 56, Pcorrected < 0.05) and the left frontal lobe (cluster size= 232 voxels, peak voxel MNI coordinate: x= -38, y= -30, z= 68, Pcorrected < 0.05). Moreover, post-training showed significantly decreased activation at both sides of the tempoparietal cortex (for the right side, cluster size= 219 voxels, peak voxel MNI coordinate: x= 64, y= -38, z= 18, Pcorrected < 0.05, for the left side, cluster size= 148 voxels, peak voxel MNI coordinate: x= -58, y= -38, z= 18, Pcorrected < 0.05) (Figure 2). Conclusions:　The working memory training could expand working memory capacity. This effect may due to increased activation of parietal and frontal cortex however decreased activation of tempoparietal cortex.

この発表はワーキングメモリの訓練に関する発表でした．ワーキングメモリの訓練の前後での活性領域の違いを検討していました．言語的な内容ではなく空間的な内容のワーキングメモリ訓練課題を用いていたため，非常に今後の研究の実験を設計する上で参考になった．
 

発表タイトル       ：3197 Abnormal dynamics of intrinsic brain functional networks in Parkinson’s disease 著者                  ： Jinhee Kim, María Díez Cirarda, Marion Criaud, Sang-Soo Cho, Alexander Mihaescu, Mikaeel Valli, Christine Ghadery, Sarah Coakeley, Antonio Strafella セッション名       ： poster Abstruct            ： Introduction: Parkinson’s disease (PD) is a neurodegenerative disorder characterized by nigrostriatal dopamine depletion leading to whole-brain neural circuit changes (1, 2). There have been previous studies showing altered functional connectivity and disrupted topological organization in PD patients using static resting-state fMRI data (rs-fMRI) (3-5) However, dynamic resting-state functional connectivity network (d-FNC) in PD remains largely unknown. In this study, we aimed to examine time-varying aspects of functional network connectivity and topological properties in PD. Methods: Thirty-one PD patients ON medication (65.7±7.2 years, 22M/9F, H&Y scale: 1.9±0.2) and twenty-four healthy controls (64.5±8.3 years, 12M/11F) were included in the analyses. Resting-state fMRI data were collected in a GE 3T scanner (TR = 2000 ms, TE = 30 ms, FA=60, FOV: 220 mm, matrix size: 64 x 64, 36 5-mm slices, durations: 8 min 24 sec). Preprocessing was performed using SPM 12 (motion correction, normalization to the MNI template, and spatial smoothing with a 6 mm Gaussian kernel). With the preprocessed rs-fMRI data, 44 independent components (ICs) were obtained using group spatial independent component analysis (6). In dynamic functional connectivity (FC) analysis based on previous studies (7, 8), time-varying covariance matrices were extracted using a sliding windows approach (TRs = 44, 1 TR steps). We estimated dynamic FC states using K-mean clustering methods, then compared the temporal properties (i.e., dwell time of state and number of state transitions) (p < .05, FDR corrected) and FC strength between groups (p < .05, FWE with 10,000 permutations). In the graph theory based approach, we calculated topological metrics (i.e., global and local efficiency)(9) of d-FNC across all the time windows and compared these total variances between groups. Results: Dynamic FC analysis revealed that compared to HC, PD patients spend significantly shorter time in state I (a state which occurs frequently and has weak FC between components) and longer time in state II (a rarer state but with stronger FC pattern). For state I, PD patients showed hypoconnectivity between visual, somatosensory, and default mode networks compared to HC. The UPDRS motor symptom score was positively correlated with the number of the state transitions in PD patients. In addition, dynamic topological analysis showed that PD patients exhibited higher variance of global efficiency than HC. Conclusions: Our findings provide new insight into understanding the role of the dynamic FC network in PD which is not as well characterized. This research will provide a better understanding behind the pathological mechanisms of PD.

 
この発表はパーキンソン病患者の脳の機能的ネットワークに関してコントロール群の健常者と比較検討していました．パーキンソン病患者および健常者のレスティングステートを計測し，スライドウィンドウによるダイナミックコネクティビティ解析を行い，k-meansにより2クラスに状態を分類していた．自分たちの研究室に近いアプローチを使っており非常に興味深いと感じた．
 

発表タイトル       ：1389 Brain Network of Emotion Regulation in Soldiers with Trauma 著者                  ： D Rangaprakash, Michael Dretsch, Thomas Daniel, Thomas Denney, Jeffrey Katz, Gopikrishna Deshpande セッション名       ： Morning Symposia Abstruct            ： Introduction: Our ability to shape our emotional experience is termed emotion regulation (ER) [1], involving voluntary modification of emotions elicited in response to exogenous stimuli. Several functional MRI activation studies have consistently identified the middle frontal gyrus (MFG), anterior cingulate and insula to be involved in it [1]. Their limitation lies in the inability to explain the interrelationship between these regions, i.e. connectivity. The brain network of ER either in healthy adults or in psychiatric disorders like posttraumatic stress disorder (PTSD) and mild-traumatic brain injury (mTBI) has been elusive. Emotion dysregulation (ED) is regarded a primary cause for many symptoms observed in PTSD and mTBI [2]. Using fMRI data collected during an ER task, we obtained the network of ER in healthy soldiers and ED in soldiers with comorbid PTSD and post-concussion syndrome (PCS, or chronic mTBI). Methods: 59 male U.S. Army soldiers were recruited (comorbid PCS+PTSD=36, combat controls=37, matched in age, race and education). FMRI data was acquired in a Siemens Verio 3T scanner (EPI sequence, TR/TE=600/30ms, flip-angle=55o, voxel size=3.5×3.5×5mm3). The ER task (Fig.1) was similar to Urry et.al. [3]. Participants were presented images eliciting a negative emotional response, and were asked to either “maintain” their emotional response, or “suppress” it (reduce negative feelings, requiring ER). There were 4 task blocks, with 24 trials in each block. Standard pre-processing was performed in SPM (realignment, smoothing [8mm kernel], normalization to MNI space). We first identified significantly activated regions during ER (see Fig.2 for region selection procedure). Hemodynamic deconvolution was performed [4] on mean time series extracted from identified regions, to minimize the non-neural intra-subject HRF variability [5]. We employed effective connectivity (EC) modeling using Granger causality (GC) [6] to assess directional causal relationships between identified regions, similar to recent works [7]. Subject-wise EC between all regions were obtained, using which the networks of ER in healthy soldiers (suppress>maintain) and its impairment in PCS+PTSD (control>PCS+PTSD for ‘suppress’ condition) were obtained (p<0.001, Bonferroni corrected) (Fig.3). We provide novel evidence for the brain networks of both ER and ED in a clinical population. Results: We investigated brain networks of ER in healthy soldiers, and ED in PCS+PTSD. We defined our ROIs around the 9 regions activated during the ER task (Fig.4). With EC analysis, we found the ER network having a top-down structure with the MFG driving the rest of the network (insula, medial prefrontal, amygdala and lateral parietal regions) (Figs 5a,5b,5c). During ED this network was imbalanced with reduced prefrontal connectivity and elevated subcortical and lateral parietal connectivity (Figs 5d,5e,5f). Our ER network fits well with prior findings [1, 8], which identified the pivotal role of MFG in the initiation of ER. MFG is implicated in executive functions like cognitive control [1], which are necessary for regulating emotions. Soldiers with PTSD exhibit impaired emotional processing [9] and impaired cognitive functions associated with the MFG [10],. All directional connections are traceable to the MFG, implying that MFG could be the source of ER [1]. As for ED, the MFG emerged as the key source of disruption in PCS+PTSD. All connections from MFG had reduced connectivity, whose “ripple-effect” culminated in disinhibition of amygdala, which might translate to symptoms like flashbacks, trauma re-experiencing and hyperarousal. This fits well with behavioral manifestations of these conditions [2]. Conclusions: In summary, we identified the MFG as pivotal to ER in healthy soldiers and ED in PCS+PTSD. Our findings are significant given that these regions are implicated in prior activation studies [1, 8], but a precise understanding of the underlying network structure and their causal relationships had not emerged so far.

この発表は感情調節における機能的接続性に関する発表でした．兵士に戦場における不快な画像を提示し，感情を制御するもしくは維持することを要求された．
被験者の兵士は健常者と感情調節不全に別れ二群の違いについて検討された．
結果の表し方が自分が依然行っていたような内容であったが，この研究では有効の接続性が使われており，今後の利用していきたいと感じた．
 

発表タイトル          ：Neural underpinnings of mutual gaze and joint attention using hyperscanning functional MRI 著者                  ： Hiroki Tanabe セッション名       ： Morning Symposia Abstruct            ： Mutual gaze provides a communicative link between humans, prompting joint attention, which is the ability to coordinate attention between interactive social partners with respect to objects or events to share an awareness of them. Joint attention is of particular importance during early social development representing the prerequisite of theory-of mind and social communication. To elucidate their neural underpinnings, we conducted several experiments employing hyperscanning functional MRI combined with online video cameras and voice exchange system. I will show the results of these studies and discuss core neural mechanisms of mutual gaze and joint attention.

この発表は2台のfMRIによるハイパースキャニングの研究でした．各fMRIにビデオカメラを設置しfMRI内で二人の共同歴な注意に関する脳活動を計測していた．一般線形モデルにおける誤差を解析しており，ほかの誰も関心を持たないような点に注目しており，素晴らしいと思った．
また実験環境にもこだわりが見られ，学ぶべきものが多くあった．

発表タイトル       ：3721 Successful encoding activation modulated by empathic traits in memory for highly empathetic people 著者                  ： Natsumi Kondo, Hikaru Sugimoto, Takashi Tsukiura セッション名       ： poster Abstruct            ： Introduction: Empathic ability is crucial in understanding intentions of others. Previous studies have demonstrated that personality traits of empathy are correlated with individual abilities of memory (Beadle et al., 2013; Wagner et al., 2015). However, little is known about the neural mechanisms underlying the empathy-memory interaction. The present fMRI study investigated encoding success activation (ESA) modulated by multiple kinds of empathic trait in memory for highly empathetic people. Methods: Twenty-four right-handed, college-aged healthy women participated in this study (mean age: 21.8, SD: 1.6). All participants were recruited from the Kyoto University community, and paid for their participation. They gave informed consent to a protocol approved by IRB of the Graduate School of Human and Environmental Studies, Kyoto University. All participants performed both encoding and retrieval tasks, and neural activation was measured only in the encoding phase. During encoding, participants were presented with pairs of an unfamiliar face and a sentence describing hypothetical action, and were required to rate how empathetic the faces presented with the hypothetical actions are. After the encoding, participants were presented with previously learned and new faces one by one, and were required to recognize whether each face was learned in the encoding phase. In addition, individual traits of empathy were evaluated by the affective and cognitive empathy (Davis, 1980; Sakurai, 1988) and the Japanese version of EQ-SQ questionnaires (D-score) (Baron-Cohen et al., 2003; Baron-Cohen & Wheelwright, 2004; Wakabayashi et al., 2006). All encoding trials were divided into highly empathetic (High) and low empathetic (Low) faces by subjective ratings during encoding, and all High and Low trials were subdivided into subsequent hits (H) and misses (M). ESA was identified by H vs. M in each condition of High and Low, and the empathy-related enhancement of ESA for face memories was identified by comparing between ESA in the High and Low conditions. In addition, we investigated correlations between the empathy-related enhancement of ESA and each empathic trait of the affective and cognitive empathy, and the D-score. All MRI data were acquired by a Siemens MAGNETOM Verio 3T MRI scanner. A gradient echo EPI sequence for functional images was employed by the following parameters (TR=2 s, TE=25 ms, flip angle=70 degree, 39 slices, 3.5 mm slice thickness). The preprocessing and statistical analyses for all functional images were performed by SPM12. Results: In behavioral data, response time (RT) during both encoding and retrieval was significantly smaller in High than in Low (Encoding: F=6.40, p<.05, ηp2=.22; Retrieval: F=7.62, p<.05, ηp2=.25). In addition, RT during the successful retrieval of highly empathetic faces was significantly smaller than that in the other conditions (F=6.18, p<.05, ηp2=.21). fMRI data in the regression analyses demonstrated that the empathy-related enhancement of ESA in a posterior part of the left dorsomedial prefrontal cortex (dmPFC) was positively correlated with individual score of the affective empathy, and that the empathy-related enhancement of ESA in an anterior part of the left dmPFC was positively correlated with individual score of the cognitive empathy. In addition, a significant correlation between the empathy-related enhancement of ESA and individual D-score was identified in the right temporoparietal junction (TPJ). Conclusions: The present findings suggest that ESA increased in highly empathetic faces could be associated with three different regions of the posterior dmPFC, anterior dmPFC, and right TPJ, each of which reflected individual difference in the affective empathy, cognitive empathy, and EQ-SQ difference (D-score). The enhancing effect on responses for highly empathetic people in memory-related processes could be modulated by several different components of empathic traits.

この発表は共感時における脳活動に関する発表でした．賦活を見た後にその領域における機能的結合を確認していた．スモールボリュームコレクションという方法で特定領域の賦活を見ることができることがわかった．今後の研究で用いる場面があれば利用したい．
 
学会参加報告書

報告者氏名	三好巧真
発表論文タイトル	脳機能ネットワークと唾液内代謝物質における数息観の影響
発表論文英タイトル	Effects of breath-counting meditation on functional brain network and salivary hormones
著者	三好巧真, 日和悟, 廣安知之
主催	Organization for Human Brain Mapping
講演会名	23rd Annual meeting of the Organization for Human Brain Mapping
会場	Vancouver Convention Centre
開催日程	2017/06/25-2017/06/29

 
 

1. 講演会の詳細

2017/06/25から2017/06/29にかけて，Vancouver Convention Centreにて開催されましたOHBM2017に参加いたしました．こOHBM2017は，Organization for Human Brain Mappingによって主催された研究会で，脳科学研究者を中心に，ヒトの脳に関する研究を行う研究者も参加して，ニューロイメージング研究の活性化を図るための議論を行うことを目的に開催されています．
私は全日程，参加いたしました．本研究室からは他に廣安先生，日和先生，M2片山さん，M2和田さん，M2萩原さん，M2吉武さん，M2石原さん，M2玉城さん，M1藤井さん，M1水野さん，M1池田さん，M1相本くん，M1石田くん，M1中村圭くんが参加しました．
 
 
 

2. 研究発表
   • 発表概要

私は27日の午後のセッション「Well-being Computing」に参加いたしました．発表の形式はポスター発表で，2時間の講演時間となっておりました．
今回の発表は，「Effects of breath-counting meditation on functional brain network and salivary hormones」です．以下に抄録を記載致します．

Introduction: Mindfulness meditation has been shown to reduce stress and improve concentration. In recent years, researches on brain activity during meditation using fMRI are extensive [1] . In addition, changes in the human condition by meditation also appear in the form of various biological information. Here, we examine changes in brain activity and the salivary hormone, cortisol, during meditation. Methods: In this study, 24 novices of meditation and 2 meditation experts participated in this experiment. All participants used Zen breath-counting meditation method (Susokukan). We measured brain activities during the meditation using fMRI in all participants. Salivary cortisol was also measured in 4 of the novices. The whole brain was divided into 116 regions using Automated Anatomical Labeling (AAL), and the correlation coefficient of the BOLD signal was calculated between each region. The correlation matrix was binarized with an edge density of 15%, and the degree centrality and betweenness centrality of each region was calculated. The resulting 232-dimensional data set, composed of the degree centrality and betweenness centrality in the 116 areas, was divided into two groups by reduced k-means clustering [2], and the features of each group were analyzed. Results: Through reduced k-means clustering, the large 232-dimensional data set was decomposed into 1-dimensional data (first principal component). All subjects were divided into cluster A including the 2 experts and cluster B consisting only of beginners. The 2 experts were located close to each other in the cluster. Fig.1 illustrates the principal component loading. The variables with a significant positive principal component loading were degree centrality of right superior frontal gyrus, medial (rSFG medial), right hippocampus, right thalamus (rTHA), and right putamen (rPUT). The rSFG medial and right Hippocampus belong to the default mode network (DMN). The DMN is one of the brain networks activated during meditation [1]. The rTHA transfers information to the basal ganglia and the rPUT controls the limbic system. On the other hand, the variables with the most significant negative principal components loading were degree centrality of right occipital lobe and right cuneus (rCUN). Right occipital lobe and rCUN are areas related to vision. This suggests that cluster A reflects a reduction in the connection between regions related to vision, and increased connection between regions within the DMN. The characteristics of betweenness centrality did not show a large difference between the two groups. Fig.2 illustrates the changes in salivary cortisol. Subjects 1 and 2 were classified within cluster A and subjects 3 and 4 were classified within cluster B. In cluster A, the degree centrality of rTHA and rPUT was high. Hormonal secretion was controlled by the limbic system, so there is a possibility that the variation of salivary cortisol in subjects 1 and 2 was small. Conclusions: In this paper, we examined changes in brain state and salivary cortisol in novice and expert meditators using fMRI. In cluster including experts, the connection between the region related to vision and the other region decreased, and the connection between the region of DMN and the other region increased. No change in salivary cortisol was observed in the same cluster. Therefore, it was suggested that meditation changes the connection of the region related to vision and DMN, and furthermore, the decreases the fluctuation of salivary cortisol.

 
 
 

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
 
・質問内容1
質問者の氏名は不明です．こちらの質問は，「機能的な結合だけでなく，構造的な結合は見てないのか」というものでした．この質問に対して私は，「今回の発表では機能的結合だけだが，構造的な結合を考慮する必要がある」と回答しました．
 
・質問内容2
質問者の氏名は不明です．こちらの質問は，「数息観とはどのような瞑想方法なのか」というものでした．この質問に対して私は，「呼吸を数えることで呼吸に注意を向ける訓練」と回答し，数息観の実践方法を説明しました．また，「瞑想の種類によって脳活動に違いはあるのか」という質問を受けました．この質問に対して私は，「集中瞑想や洞察瞑想などがあり，結合が異なることが知られている．」と回答しました．
 
・質問内容3
質問者の氏名は不明です．こちらの質問は，「瞑想時の脳活動を計測しているが，対象としてのオフセットはあるのか．」というものでした．この質問に対して私は，「今回の実験では5分間の瞑想状態と5分間の安静状態がある．」と回答しました．
 
・質問内容4
質問者の氏名は不明です．こちらの質問は，「唾液の採取は被験者間で時間を統一しているのか」というものでした．この質問に対して私は，「すべての被験者において午前10時から実験を行っているため，統一されている」と回答しました．
 

• 感想

今回の発表は，私にとって初めての国際学会でありました．英語で発表ということで，思うように伝えたいことを伝えるのが難しかったですが，相手の様子をうかがいながら話す内容を変えることが徐々にできるようになったと思います．多くの方がポスターを見に来てくださり，興味深い研究だと言っていただきました．質問は，深く踏み込んだものはありませんでしたが，参加者皆さんが脳機能研究を行っているだけあって，エビデンスが弱い部分の指摘をいただくこともできました．時に英語を聞き取れず，先生方に助けていただく場面もありましたが，充実した発表になったと感じています．今後は，さらに先行研究等を十分に参考にし，脳機能研究に求められる要件を抑えながら研究を進めていきたいと思います．
 
 

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました．
 

発表タイトル       ：　Neuronal processing of affective touch in patients with Posttraumatic Stress Disorder 著者                  ： Timmy Strauss, Kerstin Weidner, Ilona Croy セッション名       ： Disorders of the Nervous System Abstruct            ： Introduction: Posttraumatic stress disorder (PTSD) is a prevalent mental health condition triggered by exposure to actual or threatened death, serious injury or sexual assault. Furthermore PTSD is characterized clinically by hyperarousal and flashbacks. In this connection recent functional MRI-studies, in which patients were confronted with their traumatic memory (script- driven trauma imagery), prove that affected people have both an activation in right anterior Insula and a deactivation in right rostral anterior Cingulum (rACC) during flashbacks. We aimed to use a different paradigm by applying affective touch in a social versus a non-social condition. Therefore our study`s intention was to reveal neural correlates triggering hyperarousal and flashbacks in interpersonal touch. Methods: Twenty patients (19 women, aged between 24 and 58 years) with a history of sexual abuse and physical maltreatment and a diagnose of PTSD were compared to 20 age and sex matched healthy control subjects. All subjects answered questionnaires about current symptoms of PTSD, dissociative symptoms, depression and maltreatment (e.g. Psychopathy Checklist, Beck Depression Inventory II, Childhood Trauma Questionnaire, Asperger Questionnaire). Functional magnetic resonance data were acquired on a Siemens 3 Tesla scanner using a protocol with a T2*-weighted gradient-echo, echo-planar imaging sequence (TR = 3 seconds, TE 51ms, flip angle 90°, 25mmx6mm axial slices, 3.6×3.6mm in-plane resolution). In our fMRI we stroked them on their left forearm with a human hand and a brush, modifying the velocity of stroking between slow and fast. Each condition (hand slow, hand fast, brush slow, brush fast) took six minutes with a constant switch of on- and off-conditions after 15 seconds. Afterwards they had to assess both the pleasantness and the intensity of stroking. A high resolution T1 sequence (TR = 3 seconds, 0.7x1mm in-plane resolution) was added for precise anatomical mapping of functional data. The scanning planes were oriented parallel to the anterior-posterior commissure line and covered the whole brain (threshold p<0.001). Results: PTSD patients rated both hand stroking conditions as extremely unpleasant while healthy subjects enjoyed those conditions. Behavioral data was mirrored by fMRI results, showing a highly significant “group by agency” interactions effect in right primary somatosensory cortex, right hippocampus, right superior frontal and temporal gyrus as well as right posterior insula. Post hoc analysis revealed that patients had significantly higher primary somatosensory activation than controls. Furthermore, patients showed a pronounced deactivation in the right hippocampus, which was not present in the controls (figures 1+2). No major group differences were found in the non-social stroking conditions. Conclusions: Hippocampal deactivation in patients may indicate a suppression of traumatic memories triggered by touch. If so it can be expected that intimate touch in daily life triggers re- experiencing of the trauma. In this connection a causal relation remains speculative: we suppose that touch leads to an automated suppression of hippocampus which prevents the recovery of PTSD. It should be discussed whether hippocampal activation is a biomarker for disease maintenance and whether a suppression of deactivation e.g. via neurofeedback has positive therapeutic implications.

PTSDの患者に対して社会的接触を繰り返すと，逆効果であることが知られています．本研究では，社会的接触における神経処理をfMRIを用いて検討しています．記憶に関係する海馬や，情動に関連するSTGにおいて対象群と比べ変化が見られました．Amygdalaとの関連性を見ると，より患者の神経基盤がわかると考えました．
 

発表タイトル       ： Meditation, resting state connectivity, and sustained attention: An RCT in middle school children 著者                  ： Clemens Bauer, Camila Caballero, Ethan Scherer, Martin West, Susan Whitfield-Gabrieli, John Gabrieli セッション名       ： Higher Cognitive function Abstruct            ： Introduction: Mindfulness meditation describes a set of mental techniques to train attention and awareness (1). Interest in mindfulness-based approaches with adults has grown rapidly, and there is expanding research suggesting these are efficacious approaches to promoting psychological health and well-being (2). Interest has spread to applications of mindfulness-based approaches with children and adolescents, especially since children are increasingly exposed to longer periods of persistent and intensive demands to improve academic performance (3). We investigated whether mindfulness training promotes sustained attention in middle-school children, and whether such training alters functional connectivity of cortical midline structures of the default-mode network (DMN) that have been suggested to be engaged in mind wandering (4). Methods: 100 sixth-graders participated in a randomized controlled trial (RCT) at a charter school in Dorchester, MA, USA. Intervention Group (MT) received a mindfulness curriculum during their last 45 minutes of their school day, 4 times a week for 8 weeks. Active Control Group (CN) received SCRATCH computer programming. 40 children (20 MT) additionally underwent MRI scans at MIT. Before and after the intervention, we measured attention by the Sustained Attention Response Task (SART)(5) and DMN connectivity by 5 min resting state scans. 32 children (Female: 20, mean age 12.24 years (SD 0.40)) were included after excluding 8 participants (>2 mm displacement). We used multivariate regressions controlling for baseline, age, IQ and gender to determine beta coefficients of the treatment effect. Statistical analysis was performed in R and Connectivity analysis using the Conn Toolbox (6). All analysis are non-parametric cluster-size FDR < 0.05 corrected. Results: SART: There was a significant difference in SART performance between the groups after the intervention on sustained attention as measured by accuracy on “Go” trials (b = 0.076, t(19) = 2.26, p = .036) (Fig.1). This reflected equal performance for the two groups before intervention, with the MT group improving and the CN group declining after intervention. Brain Connectivity: Mixed within-between-subject 2×2 ANOVA showed a reduction in connectivity at post-intervention for MT but not for CN between DMN and frontoparietal network (FPN) including the rDLPFC (BA 9) and SMA (BA 6). MT group significantly predicted reduced connectivity z-scores between DMN and rDLPFC (b = -0.22, t(26) = -3.613, p = .001, Fig 2a). Change in “Go” accuracy (ΔGo) significantly correlated with change of connectivity between DMN & BA 6 (ΔDMN/BA6)(Fig.2b). Conclusions: Mindfulness training in middle school children can improve sustained attention and reduce functional connectivity between cortical midline structures of the DMN & FPN, which has been associated with greater cognitive flexibility (7). If these results are replicated in future studies, one might consider including mindfulness training in the curriculum of schools.

この発表は学生に対するマインドフルネスの8週間のプログラムの効果に関するものでした．瞑想群とコントロール群を用意し，瞑想群に8週間の瞑想訓練を行ったところ，瞑想群は持続的な注意が向上し，DMNの結合が減り，ストレスの低下にもつながったという結果が得られました．この結果は，瞑想訓練が効果的であることを示し，1日の瞑想訓練でも効果があるのか，この研究によって明らかにされたネットワークに着目し，検討する必要があると考えました．
 

発表タイトル       ： Detecting mindfulness state from MEG/EEG functional connectivity 著者                  ： Alexander Zhigalov, Erkka Heinilä, Tiina Parvianen, Lauri Parkkonen, Aapo Hyvärinen セッション名       ： Imaging Methods Abstruct            ： Introduction: Mindfulness meditation involves sustaining attention towards a selected object (e.g., sensation of respiration) and away from external or internal sources of distraction (Chow et al., 2016). Prior research shows that mindfulness training reduces stress, increasing both physical and mental well-being, and is a useful intervention for patients (Tang et al., 2015). Since sustained attention is difficult to perform, neurofeedback devices have been developed to help sustaining attention, for instance, by facilitating the maintenance of high EEG alpha amplitude (Chow et al., 2016). An alternative kind of feedback could be provided by the detection of wandering thoughts that involuntarily interrupt sustained attention. In this study, we developed a novel MEG/EEG functional connectivity-based classification approach that aims at detecting mind wandering in real-time and that could be applied in neurofeedback. Methods: We analyzed offline MEG data recorded from ten healthy subjects. Two identical experimental sessions were carried out with interval of one week, and each session consisted of four consequent tasks in an order balanced across sessions. The tasks were: resting (3 min), mindfulness meditation (6 min), planning the future (4 min), and evoking negative emotions (4 min). The instructions for switching between tasks were displayed on a monitor screen. In this analysis, we compared neuronal activity during rest against the three other conditions using two classification algorithms. The MEG recordings were divided into 2-s epochs with 75% overlap, and the epochs were labelled according to the condition. First, we applied the spectral linear discriminant analysis (LDA; Kauppi et al., 2013; available in the “Spedebox” package). Essentially, the method computes independent components in frequency domain (Hyvärinen et al., 2010) and then applies LDA. Second, we developed an alternative connectivity LDA algorithm that performs independent components analysis (Hyvärinen, 1999) in time domain, computes connectivity metrics (cross-frequency coupling; Canolty and Knight, 2010) between the independent components and finally applies LDA. The performance of the algorithms was compared at the subject level using Wilcoxon rank sum test. Results: The mean classification accuracy for both classifiers was well above the chance level (0.5). For rest vs. mindfulness conditions, the spectral LDA provided a mean classification accuracy of 0.68±0.011 (mean±SEM), while connectivity LDA showed significantly (p < 0.001, Wilcoxon rank sum) higher accuracy 0.74±0.015. Similarly, the mean accuracy for spectral LDA (0.64±0.014) was smaller (p < 0.004) than the accuracy for connectivity LDA (0.69±0.009) in the rest vs. future planning conditions. Again, when classifying rest vs. negative emotions conditions, the connectivity LDA (0.70±0.011) outperformed (p < 0.006) the spectral LDA (0.64±0.011). Conclusions: Our results show that detection of mind wandering is possible by applying machine learning on MEG data, at least in the basic offline setting. We further showed that connectivity LDA provides a better classification accuracy compared to the spectral LDA, suggesting that functional connectivity is more sensitive for detecting mind wandering. In the future, the MEG/EEG connectivity based real-time neurofeedback may open novel avenues for both examining the functional role of connectivity between different brain areas and frequency bands in healthy subjects and for developing novel therapeutic approaches for brain disorders associated with attention impairment.

この発表では，マインドフルネス瞑想の支援のためのニューロフィードバックに関する発表が行われていました．EEGをもちいて，瞑想状態を定量化するだけでなく，どのようにリアルタイムにフィードバックするのかについても機械学習を用いて検討されていました．自分の研究においても，フィードバック方法を頭において研究をしていく必要があると考えました．
 
 
 

発表タイトル       ： Can brain state be manipulated to emphasize individual differences in functional connectivity? 著者                  ： Emily Finn, Dustin Scheinost, Daniel Finn, Xilin Shen, Xenophon Papademetris, R Constable セッション名       ： Modeling and Analysis Methods Abstruct            ： Introduction: While neuroimaging studies typically collapse data from many subjects, brain functional organization varies between individuals, and characterizing this variability is crucial for relating brain activity to behavioral phenotypes. Rest has become the default state for probing individual differences, chiefly because it is easy to acquire and a supposed neutral backdrop. However, the assumption that rest is the optimal condition for individual differences research is largely untested. In fact, other brain states may afford a better ratio of within- to between-subject variability, facilitating biomarker discovery. Here, we present analyses of within- and between-subject variability across a wide range of scan conditions to determine if and how brain state can be manipulated to emphasize individual differences in functional connectivity. Methods: Data were obtained from the Human Connectome Project (Van Essen et al., 2013), 900 subjects release. Analyses were limited to 716 subjects that had complete data for each of nine functional scans: EMOTION, GAMBLING, LANGUAGE, MOTOR, RELATIONAL, REST1, REST2, SOCIAL and WORKING MEMORY (WM). Using a 268-node functional brain atlas, for each subject, we calculated nine connectivity matrices consisting of the pairwise correlation coefficients between each possible pair of nodes using data from each scan condition, respectively. Because connectivity matrices are symmetric, we extracted the unique elements by taking the upper triangle of the matrix; this results in a 1×35,778 vector of edge values for each subject for each condition. These vectors can then be compared using Pearson correlation either between different subjects in the same condition (yielding a 716 x 716 between-subject correlation matrix for each condition, with 255,970 unique values representing similarity between all possible subject pairs), or within the same subject across conditions (yielding a single 9 x 9 within-subject correlation matrix for each subject). Results: Our analysis showed that brain state does affect between-subject variability. The RELATIONAL task had the highest between-subject similarity (r = 0.53), while the two REST sessions, along with the MOTOR session, had the lowest between-subject similarity (r = 0.35). Given equal scan durations, two different tasks sometimes showed higher within-subject similarity than the two rest scans. Mean similarity between the RELATIONAL condition and the EMOTION, GAMBLING and WM conditions (r = 0.62-0.64) all exceeded mean similarity between REST1 and REST2 (r = 0.55). We also replicated the identification experiments described in Finn et al. (2015), in which a target matrix from one scan condition was used to identify the same individual from a set of matrices from a different scan condition. While rates were well above chance for all condition pairs, some pairs were more successful than others (accuracy range = 15%-92%). Interestingly, conditions that made subjects look more similar to one another tended to make better databases for identification experiments (r = 0.82, p = 0.007). Conclusions: We present these observations as proof-of-principle that individual differences in functional connectivity do, in fact, depend on the condition in which they are measured. We hope these results provide a jumping-off point for more detailed investigations into how brain state affects both within- and between-subject variability, which will help determine which conditions are optimal for individual differences research. Depending on the trait or behavior under study, certain tasks may bring out meaningful idiosyncrasies across subjects, essentially enhancing the individual signal in networks of interest beyond what can be measured at rest.

この発表で着目したのは，resting-stateが本当に個人差の無いニュートラルな状態なのかということに問題にしていたからです．確かに，resting-stateが個人間で大きくばらつく場合，実験設計を見直す必要があります．個人差のない状態を検証することは今後の研究において必要なことであると感じました．
 
 

発表タイトル       ：Challenges in measuring individual differences in fMRI functional connectivity in healthy aging. 著者                  ： Linda Geerligs, Kamen Tsvetanov, . Cam-CAN, Richard Henson セッション名       ： Modeling and Analysis Methods Abstruct            ： Introduction: Many studies report individual differences in functional connectivity (FC), such as those related to age. There are a number of factors which can affect FC results in an aging sample, such as different contributions of the vascular component of the fMRI signal (Murphy et al., 2013; Tsvetanov et al., 2015) and effects of head motion (D’Esposito et al., 1999; Geerligs et al., 2015a). Most studies attempt to correct for these and other confounds by using a range of pre-and post-processing techniques. However, large discrepancies between the results of different studies (e.g. Betzel et al., 2014; Chou et al., 2013; Ferreira et al., 2015; Geerligs et al., 2015b) suggest that these analysis choices may have a big impact on the results. In the present study, we systematically explore a number of important confounds and the effects of different methods to address them, using two resting-state fMRI sessions from a large sample of adults uniformly spread across the adult lifespan. Methods: Two hundred and thirty-six participants (18-88 years old, M = 53.8, SD = 17.8, 119 males and 117 females) were included in this study, from the population-based sample of the Cambridge Centre for Ageing and Neuroscience (CamCAN). Eyes-closed resting state functional magnetic resonance imaging (fMRI) data were collected in two scanning sessions, which were three months to three years apart. We analysed the data using different pre- and post-processing pipelines. We varied which nuisance regressors were used, such as motion, cerebrospinal fluid (CSF), white matter (WM) and vascular signals, and which filters were applied (band-pass or high-pass filters, with or without pre-whitening). A range of outcome measures were used, such as the relation between FC and head motion and vascular health, as well as various reliability indices. Results: We observed a strong age-related decline in vascular health (r=-0.50, p<0.001), which partly mediated the age-related decline in mean functional connectivity, even after motion, CSF and WM regression. Additional analyses revealed that regression of CSF and WM signals has a marked effect on the distribution of age-related changes in FC (see figure 1): Prior to any signal regression, both negative and positive associations were found between age and FC, but after CSF or WM regression, there were almost no positive associations between age and connectivity. These results suggest that the association between FC and vascular health is due to the presence of residual global signals, which are most likely to have a physiological, rather than neural, origin. To deal with these remaining global signals, we created a new regressor based on fMRI signal in blood vessels. We found that this regressor largely accounted for these vascular health effects, especially when combined with regression of mean connectivity over participants (see figure 2 A-C). Additional nuisance regression steps led to improved reliability of the connectivity matrices. Furthermore, band-pass filtering, as compared to high pass filtering only, worsened the reliability of connectivity matrices (see figure 2 D-F). Conclusions: Together, these results suggest that individual variations in vascular health, and pre-processing choices, have a strong effect on estimates of mean connectivity. When these differences in mean connectivity are not accounted for, incorrect conclusions might be drawn about the effects of aging on FC. Effects of vascular health can be reduced by including an additional vascular signal regressor, which captures physiological components of the global signal. We advocate including this vascular signal in addition to motion parameters, CSF and WM signals, as this is likely to produce more reliable results that are less affected by vascular health. In addition, we propose that it is more appropriate to focus on the relative pattern of age-related changes across ROIs, by applying normalisation methods like regression of mean connectivity across participants.

この発表で着目したのは，前処理が及ぼす結果の違いを検討している点です．この研究では，前処理の選択によって，誤った結果が導かれる可能性があるということを示しています．この結果から，自分の研究においても前処理が適切であるかの検討を十分に行う必要があることを感じました．
学会参加報告書

報告者氏名	吉武　沙規
発表論文タイトル	最適化HRFによるfNIRSデータ解析
発表論文英タイトル	Adaptive HRF analysis of fNIRS data
著者	吉武　沙規
主催	医療情報システム研究室
講演会名	OHBM2017
会場	Vancouver Convention Centre
開催日程	2017/06/25-2017/06/29

 
 

1. 講演会の詳細

2017/06/25から2017/06/29にかけて，Vancouver Convention Centreにて開催されましたOHBM2017に参加いたしました．この学会は，Organization of Human Brain Mapping(OHBM)によって主催された学会で，この学会は，神経イメージングを用いた脳機能の解明を目的として開催されています．
本研究室から他に，廣安先生，日和先生，M2の石原さん，萩原さん，玉城さん，和田さん，片山さんが，M1から池田さん，石田翔也さん，藤井さん，水野さん，三好さん，相本さん，中村さんが参加しました．
 

2. 研究発表
   • 発表概要

私は29日の12:45からのPoster Session: Poster Numbers #3000-4261に参加いたしました．ポスター前での議論が行われました．
今回の発表は，fNIRSによる計測データの解析方法についての発表を行いました．以下に抄録を記載致します．

1.    Introduction fNIRS is one of noninvasive brain function imaging devices. Activation of brain function is examined by measurement of blood flow change.　The deterministic method of activation judgment does not exist. However, GLM of the hemodynamic response function (HRF) obtained from experimental design, and the experimental data is widely used. The HRF is created by convolution of cHRF. However, the shape of the cHRF, the pick timing will also change depending on the site, person, and time. Thus, the stimulus derived from the experiment may have different magnitudes of stimulus, and in some cases stimuli that are not supposed to have been existed. In this research, we propose a method to estimate stimulus magnitude and cHRF parameters from experimental data of fNIRS. By doing this, the stimulus assumed at the time of practical design is examined. Also, due to the peak time of cHRF, temporal transmission of brain function is confirmed.   2.  Methods In the proposed method, firstly, timing as a stimulus candidate is determined. Usually, only the target stimulus is set from the actual design. In this method, stimulus candidates are determined besides stimulation so as to be the target. This stimulus candidate typing has a weighted value of 0.0 to 1.0, respectively. A group of values heavier than the stimulus candidate is called a stimulus vector. cHRF has three kinds of parameters, the first peak arrival time τ p, the later peak arrival time τ u, and the respective peak value A. HRF is configured because it has an exciting feeling with cHRF having a default value. Regression analysis of experimental data of HRF and fNIRS is carried out to obtain similarity. The weight of the stimulus vector is optimized so that the degree of similarity is highest. On the other hand, if the weight of the stimulus not targeted is large, it means that an unexpected stimulus is occurring.Using the optimized vector of stimuli, the cHRF parameters, τ p, τ u, and A are optimized so that the similarity between HRF and fNIRS experimental data is high. This operation determines stimulant and cHRF parameters for each subject and site. By examining the parameters of cHRF, propagation of brain functions and the similarities are examined.Since the effectiveness of the proposed method is considered, the proposed method is applied to the blood flow change data for the n-back task. In this experiment, n = 2 and 3, and in this experiments, the period of task and rest was 30 seconds. The total of 10 tasks was performed.During the task, images of A to E were displayed around 2.5 seconds. In the list, X was displayed around 2.5 seconds.The stimulus candidate timing was set as 2.5 seconds, and a stimulation vector was created. Stimulation candidate timing existed not only at the time of the task but also at rest.         3.  Result Fig. 1 shows the magnitude of the stimulus given to the subjects. It is confirmed that the magnitude of the stimulus presented in the task period is not uniform. The value of the stimulus vector during the 2-back task is large. Therefore, it is considered that changes in blood flow due to stimulus presentation occurred. On the other hand, the values of the 3-back stimulus vector were similar values in both the rest period and the task period. Therefore, in 3-back, due to the influence on the task during the task, the blood flow change also occurs during the rest period. In this way, the experiment is examined by optimizing the stimulus vector by the proposed method. Also, τp of the obtained HRF parameter was examined. In 3-back, the last active site was in the prefrontal cortex and the inferior frontal gyrus. This suggests that character recognition and storage are performed once again at the end of the task.   4.  Conclusion In this research, we propose a method to optimize stimulus vector and cHRF parameters from fNIRS data. This method creates a blood flow change model for each subject and position. The effectiveness of the proposed method was examined by applying n-back task.

 

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
・質問内容1
最適化を行っているstimulus magnitudeとは何なのかという質問をいただきました．私は実験を行った際に被験者が受け取った刺激の大きさであると説明しました．また，従来の手法では被験者が受け取る刺激が一定であると仮定されていること，しかし実際の計測では被験者の状態などで受け取る刺激は様々であるため，解析では実際に受け取った刺激の大きさを考慮する必要性があることを説明しました．
 
・質問内容2
なぜ，刺激の大きさを変化せせる必要があるのかという質問をいただきました．これに関しては質問1と同様に従来の手法では被験者が受け取る刺激が一定であると仮定されていること，しかし実際の計測では被験者の状態などで受け取る刺激は様々であるため，解析では実際に受け取った刺激の大きさを考慮する必要性があることを説明しました．
・質問内容3
研究内容は何なのか．なぜこの実験課題を選んだのかという質問をいただきました．研究内容については，ポスターを用いながら説明を行いました．また，実験課題については，手法の検討を行っているため，先行研究が行われており，計測の行いやすい前頭部の脳活動を見たいためという説明をしました．
 
・質問内容4
HRFを使用しているのか．結果として，レストにも刺激が見られるのか．課題は何なのか．後頭部を見ているのかという質問をいただきました．HRFは使用していること，最適化を行った結果，レスト時にも被験者が何かしらの刺激を受けていることが分かったということを説明しました．また課題はN-back課題であること，後頭部も検討していることを述べました．
 
・質問内容5
HRFの最適化はどのように行っているのかという質問をいただきました．私は，HRFの最適化の際には，最適化された刺激を固定して用いていること，回帰分析を行いt値が最大になるときのパラメータを求めてHRFの最適化を行っていることを説明しました．
 
・質問内容6
用いているHRFはfMRIで使用されているHRFと同じなのかという質問をいただきました．使用しているHRFはどの装置でも変わらず使用されている血流動態反応を表す関数であるので，同じであることを述べました．
 
・質問内容7
方法を説明と，レストについて尋ねられました．方法についてはポスターを用いて一通り説明しました．レストについては，実験設計と最適化の結果レスト時間でも刺激を確認することができたことを説明しました．
 
・質問内容8
この研究の新しい点と，GLMで用いられている血流変化モデルを使用しているのかという質問をいただきました．新しい点は，血流変化モデルの最適化を行うことで，刺激の大きさの変化やレスト中にも刺激が確認できるようになったこと，その大きさが異なる刺激1つずつで発生しているHRFを求めることができ，パラメータを用いて解析ができるようになったことを説明しました．血流変化モデルはGLMで用いられているような，刺激とHRFを畳み込むことでできるモデルと同じものを使用していることを説明しました．
 
 
・質問内容9
研究結果の見方について質問をいただきました．結果についての質問でしたが，解析内容についても知ってもらわなければわからない内容だったので，研究内容の説明を一通り行いました．また，動きのある実験でのNIRS計測データでは使用できないのではないかという意見をもらいました．動作中の心拍数も計測して，それらを考慮した解析を行えるようにするとより使いやすくなりそうだという話をしていただきました．
 
・質問内容10
研究内容の説明と課題について教えてほしいといっていただきました．研究内容については前回と同じようにポスターを用いて説明しました．また，課題についてはN-back課題であることや実験設計について説明しました．
 
・質問内容11
解析方法とGLMはfMRIで使用しているものと同じなのか，HRFは一般的なものであるのか，刺激間隔はどうなっているのか，1刺激に対して1つのHRFがあるのかという質問をいただきました．解析方法については同じようにポスターを用いて説明をしました．GLMはfMRIで使用しているものと同じように実験における刺激とHRFを用いていることを説明しました．使用しているHRFもfMRIで用いられているような一般的なものを使用していることを述べました．また1刺激に対して１つのHRFを用いているのかという質問に対しては，解析方法について説明する際に説明しました．
 
・質問内容12
まず，研究内容の説明を求められました．またこの方法ではより賦活するように見ているのではないかという質問をいただきました．研究内容の説明については同様にポスターを用いて説明しました．この方法では，より賦活するように見ているのではないかという質問に対しては，この研究では何を行っているのかもう一度説明し，賦活を検討しているのではなく，あくまで最適化を行った結果を用いて解析を行うことを説明しました．
 

• 感想

前回の国際学会での経験もあり，英語についての不安は大きくは感じていませんでした．よって今回はより自分の研究内容について知ってもらうことを目標にしていました．実際の発表では，GLMやNIRSについてなどあらかじめ知っている方に多くききに来ていただき，内容について知ってもらうことができたと考えています．しかし，内容について知ってもらうことはできましたが，より深い議論ができなかったので，より英語での会話を上達させ複数人との議論ができたほうがいいと感じました．
 

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました．
 

発表タイトル          ：　Impact of Analysis Software on Replication of fMRI Studies 著者                  ：  Alexander Bowring, Thomas Nichols, Camille MAUMET セッション名       ： Poster session Abstruct            ：Researchers now have a wide selection of tools available to process and model fMRI data. However, this ‘methodological plurality’ [1] comes with a drawback. While conceptually similar, two different analysis pipelines applied on the same dataset will not provide the same results. Differences in methods, implementations across software packages, and even operating systems [2] or software versions [3] all contribute to this variability. Compounded by a lack of data sharing, an alarming consequence of this is that most findings in the neuroimaging literature are unable to be independently reproduced. Here we explore reproducibility across neuroimaging software packages, reanalysing the data of a published neuroimaging study using SPM [6] and FSL [4]. This work is part of a larger effort to replicate a number of studies with a suite of different analysis tools, facilitated by the NIDM-Results standard for representing neuroimaging results in a software-independent fashion. We also compare the results obtained between software packages and the original publication．

この発表では，研究室内で使用しているSPMやFSLなどのニューロイメージングソフトウェアパッケージ全体での再現性を研究していました．発表内で，それぞれのソフトウェアでの結果の相違の要因も推察しており，今後はこれらの要因に注意を向けながら研究を進めていくことも考慮しなければならないと感じた．また，基本的な再現性の重要性を感じることができるポスターだったと感じた．
 

発表タイトル          ： Decoding Cortical Activity with Variational Autoencoder Supports Direct Visual Reconstruction 著者                  ： Kuan Han, Haiguang Wen, Junxing Shi, Kun-Han Lu, Zhongming Liu セッション名       ： Poster Session Abstruct            ： Humans understand and explore environments without supervision. A computational account [1] for this ability is that the brain codes internal states from which perception and action emerge with a generative model (Fig. 1A). This notion is in line with variational autoencoder (VAE) in machine learning, where latent variables represent and reconstruct data through artificial neural networks [2]. This analogy inspired us to use the VAE as a model of the human visual cortex to predict cortical fMRI responses given a movie stimulus (encoding), and reconstruct the movie from the measured responses (decoding) [3].

人間の感覚による環境の理解は，脳内での生成モデルにより近くと行動が出現する内部状態をコード化していた．この考えは人工ニューラルネットワークを用いてデータを表現し再構成するVAEと一致している．よってVAEを人間の視覚野のモデルとして使用し，研究を行っていた．結果として，皮質が感覚入力を行っており，知覚を生成するために自己組織化する推論マシンであると考えていた．私は，直接脳機能の研究を行っていないため，詳しい内容まで理解することはできなかったが，ニューラルネットワークを用いて脳機能の推論を行っていく研究そのものはとても興味深かった．また，今までとは異なる方法での検討であるため，新たな発見があるのではないかと感じている．
 

発表タイトル          ：Enhancement of empathy for pain by vicarious reward measured with skin conductance response 著者                  ：　Mizuki Nakajima, Aziem Abdullah1, Sotaro Shimada セッション名           ：Poster Session Abstruct ： Vicarious reward is a reward received by watching likable others obtaining a positive outcome. Several studies have suggested the correlation between vicarious reward and the sense of unity with the others [1, 2]. However, it has not been fully examined whether and how vicarious reward enhances the sense of unity. In this study, we investigated whether the degree of empathy for other’s pain, which was measured as the skin conductance response (SCR), was modulated by the amount of the vicarious reward received beforehand.

この研究では，代理報酬による統一意識の調整について研究を行っていた．実験では，「痛い」もしくは「楽しい」映像を繰り返し見せた後でストップウォッチゲームとして5秒±0.05秒以内に収めるゲームの成功率の異なる映像を見た．その後再び「痛み」のムービーを見てもらっている．実験中の皮膚コンダクタンス反応（SCR）を計測し，検討していた．成功率の高い映像をみた後の「痛み」の映像では，SCRは成功率の低いものよりも有意に大きく，より痛みを共感していると考えられた．この発表では，私たちが日常で感じている代理報酬に注目しておりとても興味を持つことができた．しかし，計測しているデータがSCRのみであったため説得力が足りないのではないかと感じた．しかし，今後は脳波の計測を行っていくそうなので，その結果も見てみたいと感じた．

発表タイトル          ： Fractionating frontoparietal brain networks using neuroadaptive Bayesian optimization 著者                  ： Romy Lorenz, Ines Violante, Ricardo Monti, Giovanni Montana, Adam Hampshire, Robert Leech セッション名           ： ORAL SESSION: Higher Cognitive Functions Abstruct            ：FMRI studies suggest that high-level cognitive tasks recruit a combination of spatially overlapping yet distinct frontoparietal networks [1–4]. However, understanding the exact functional role of these networks remains a challenge as the same network can be activated by inherently different tasks, e.g. [5]. The cost and difficulty of data acquisition with fMRI necessitates testing only a small subset of possible tasks. This is problematic as it can lead to misleadingly narrow functions being ascribed to a network that in reality has a broader role [3,6,7]. In the context of this many-to-many-problem, fully understanding the functional role of brain networks requires a more holistic approach that considers how brain activity changes in various task contexts [6]. While meta-analyses provide answers related to broad cognitive domains, they cannot extract information about finer-grained states [8]. Here we present a neuroadaptive closed-loop framework that combines real-time fMRI and Bayesian optimization to efficiently search across a many different cognitive tasks with the aim to optimally segregate two important frontoparietal networks. The results of this analysis were subsequently fed into a second stage of optimization, to fine-tune the parameters of the optimal tasks, in order to gain further insights into the functional roles of these networks.

この研究では，fMRIでのネットワーク解析について研究している．メタアナリシスでは，広範囲での認知領域での検討ができるが，詳細な状態に関する情報を出すことはできない．そこで，この研究では，リアルタイムfMRIとベイジアン最適化を用いて神経系ネットワークを最適に分離する方法を検討している．結果的に，ベイジアン最適化は脳の認知関係を検討する効率的な方法であると示していた．以前ベイズを用いたfNIRSにおける集団解析について勉強したことがあったので，もう一度見直してみようと考えた．この手法自体もfNIRSの解析に有用なものはないか考えてみたい．
 
 
 
 

発表タイトル          ：Noise-induced nonlinear neural dynamics as an individual trait 著者                  ： Keiichi Kitajo, Takumi Sase, Yoko Mizuno, Hiromichi Suetani セッション名           ： Poster Session Abstruct ：The human brain is a nonlinear dynamical system, which is composed of a huge number of nonlinear elements. It is known that spikes of a single neuron responding to a repeatedly presented identical noisy input show highly consistent temporal patterns across different trials (Mainen and Sejnowski, 1995). From a nonlinear dynamical systems viewpoint, this phenomenon is called “consistency” of output responses, which is defined as the reproducibility of response waveforms of a nonlinear dynamical system driven by the same input signal. This phenomenon is non-trivial because a nonlinear system starting from different initial conditions show consistent outputs after a transient period as has been observed in laser systems (Uchida et al. 2004). In the current study, we investigated how and to what degree macroscopic neural signals such as electroencephalography (EEG) exhibit “consistency” to noisy visual inputs on an individual basis.

非線形の力学系である人の脳では，反復して提示されるノイズの多い入力に応答する単一のニューロンのスパイクは一貫した時間的パターンを示すことが知られている．よってこの研究では，脳波のような巨視的な神経信号が視覚的ノイズに対してどのような一貫性を示すのかを検討していた．結果として，同一のノイズの多い視覚入力に対して個人ごとに一貫性のある応答を示していた．またこの結果は個人間の応答出力の差異は脳の違いによるものであると考察していた．よって，脳の違いは個人によって異なるため，同一のノイズに対する応答の違いから個人の識別が可能になると考えていた．私はこの研究から，現在多用されている指紋や虹彩による個人識別に新たな手法が加わるのではないかと思い，興味を持った．また，なぜこのような違いが生まれるのかについても研究すると面白いのではないかと考えた．
 
参考文献
(1)S. Tsujimoto, T. Yamamoto, H. Kawaguchi, H. Koizumi and T. Sawaguchi, “Prefrontal cortical activation associated with working memory in adults and preschool children: an event-related optical topography study,” Neuroimage, vol. 1, no. 21, pp. 283–290, 2004
(2)M. Hofmann, M. Herrmann, I. Dan, H. Obrig, M. Conrad, L. Kuchinke, A. Jacobs and A. Fallgatter, “Differential activation of frontal and parietal regions during visual word recognition: an optical topography study,” Neuroimage, vol. 3, no. 40, pp.1340–1349, 2008
(3)T. Sano, D. Tsuzuki, I. Dan, H. Dan, H. Yokota, K. Oguro and E. Watanabe, “Adaptive hemodynamic response function to optimize differential temporal information of hemoglobin signals in functional near-infrared spectroscopy,” Complex Medical Engineering (CME), vol. 1, no. 1, pp. 788–792, 2012
(4)I. Dan, T. Sano, H. Dan and E. Watanabe, “Optimizing the general linear model for functional near-infrared spectroscopy: an adaptive hemodynamic response function approach,” Neurophoton, vol. 1, no. 1, pp. 015004–015004, 2014
学会参加報告書

報告者氏名	藤井聖香
発表論文タイトル	fNIRSによる数息観中の前頭部脳活動
発表論文英タイトル	Frontal lobe activity during breath-counting meditation: fNIRS study
著者	藤井聖香,日和悟, 廣安知之
主催	Organization for Human Brain Mapping
講演会名	23rd Annual Meeting of the Organization for Human Brain Mapping
会場	Vancouver Convention Centre
開催日程	2017/06/25-2017/06/29

 
 

1. 講演会の詳細

2017/06/25から2017/06/29にかけて，カナダのVancouver Convention Centreにて開催されました23^rd Annual Meeting of the Organization for Human Brain Mappingに参加いたしました．この23^rd Annual Meeting of the OHBMは，Organization for Human Brain Mappingによって主催された学会で，学生や研究者，企業が参加して，世界中のヒト脳機能を研究している方が自身の研究を発表し，お互いの研究を議論するすることを目的に開催されています．
私は全日参加いたしました．本研究室からは他に廣安先生，日和先生，M2の石原さん，片山さん，吉武さん，萩原さん，和田さん，玉城さん，M1の三好，池田，水野，相本，石田翔也，中村圭佑，藤井が参加しました．
 

• 発表概要

私は26日の午後のポスターセッションに参加いたしました．発表の形式はポスター発表で，2時間の発表時間となっておりました．
今回の発表は，数息観中の前頭部脳活動を測定し，初心者に瞑想の出来をフィードバックする手法を提案するという発表内容でした．以下に抄録を記載致します．

【Introduction】 Mindfulness meditation is used as one of means to realize a state of mindfulness. It is attracting attention because it has the effects of stress reduction and concentration improvement. In this paper, brain activity during meditation was investigated using functional Near-Infrared Spectroscopy (fNIRS) which can measure in the condition close to everyday environments. In many previous studies, experienced meditators have been studied [1], in this study we examined how meditation beginners change their brain states by meditation.   【Methods】 The frontal lobe activity during breath-counting meditation of 19 subjects (22.15 +/- 0.65 years; 9 females) was measured by 16-channel fNIRS (Spectratech, OEG-16). Breath-counting meditation is a simple meditation method to count your own breath, even beginners can do it easily. All measurement channels of fNIRS were associated with brain regions based on Automated Anatomical Labeling (AAL). In addition, fractional amplitude of low-frequency fluctuation (fALFF) [2] which is an index of local spontaneous brain activity is calculated from time series data of cerebral blood flow change obtained in each channel, and it is transformed to Z-score (zfALFF) to compare between subjects. The zfALFF of each channel was averaged within the associated brain region. All subjects were divided into several groups by Ward’s hierarchical clustering in zfALFF of all the regions.   【Results】 Among the four clusters obtained from the hierarchical clustering, Cluster A (12) and Cluster B (4) with large numbers of subjects were examined. In Cluster A, zfALFF at right superior frontal gyrus (AAL4) was significantly higher during meditation than the resting state (p <0.05). Furthermore, the value of zfALFF during meditation was significantly lower in the left middle frontal gyrus (AAL7) (p < 0.05). Fig.1 shows the regions where significant difference was observed between the meditation and resting states of Cluster A. Additionally, because the number of samples in cluster B is small, the two regions where significant differences were observed in Cluster A were considered. The zfALFF of the right superior frontal gyrus of the four subjects of Cluster B declined from resting to meditation state. In addition, zfALFF in left middle frontal gyrus increased from resting to meditation state. Previous study [1] has reported that during meditation dorsolateral prefrontal cortex (DLPFC) was active in attention diversion and maintenance. Furthermore, it has been reported that the right DLPFC showed higher activation than the left regions in the GO/NO-GO task used to measure reaction suppression [3]. In other words, the activity of the right DLPFC during meditation is estimated to show attention to only breathing, indicating that attention to other things has been suppressed. Therefore, it is suggested that subjects in Cluster A sustained attention to their breathing because right superior frontal gyrus which was included DLPFC was activated during meditation. On the other hand, in Cluster B, the activity of the right superior frontal gyrus was reduced during meditation, and the activities of the left medial frontal gyrus were increased. Therefore, it is suggested that Cluster B did not draw attention to breathing during meditation, as brain activity of Cluster B are counter to that of Cluster A. 【Conclusions】 In this paper, we compared frontal lobe activity of meditation beginners during resting and meditation states using the fNIRS. As an indicator of local spontaneous brain activity, zfALFF was calculated. As a result of zfALFF clustering, a group with significantly higher zfALFF on right superior frontal gyrus during meditation was extracted. Because this region includes DLPFC, which has been associated with meditation in previous studies, it was suggested that 12 subjects in this group were attentive to their breathing during meditation.

 

• 質疑応答

今回のポスター発表では，以下のような質疑を受けました．
 
・質問内容1
理研所属の佐瀬さんからの質問です．こちらの質問は距離に対してMDS（多次元尺度構成法）を使用してはどうかというものでした．この質問に対する私の回答は，MDSを知らなかったため，その時は回答できませんでした．あとで調べたところ，MDSは距離行列から低次元に視覚化できるため，熟練者との距離をよりわかりやすく可視化できるのでは？というアドバイスであることがわかりました．
 
・質問内容2
質問者の氏名を控え損ねてしまいました．こちらの質問はどのようにチャンネルと脳部位の関連付けはどのようにしているのか？という質問でした．私はこの質問に対する準備をしていなかったため，日和先生にお助けをしていただきました．3Dデジタイザーを使って座標を測定し，NIRS-SPMを使っていることをお伝えしました．
 

• 感想

ポスター発表は2度目ですが，初めての国際学会ということで非常に緊張しました．自分から話しかけることが難しく，見に来てくださった殆どの人に「Do you know mindfulness?」 か 「Please any question.」のどちらかでしか話しかけられませんでした．一通りの説明は練習をしていたのでできたものの，なかなか質問を聞き取れなかったため，質問に応えられなかった場面も多々有りました．しかし，予想よりも多くの人がきてくださり，私の研究を「Nice!」と言ってくださった方もいらっしゃり，とても自信につながりました．次回はリスニング力を鍛えて，質問にも対応できるよう，練習したいと思います．
 

2. 聴講

今回の講演会では，下記の5件の発表を聴講しました．
 

発表タイトル       ：　Brain Growth and the Development of Face Recognition 著者                  ： Kalanit Grill-Spector, PhD セッション名       ： Keynote Lecture Abstract            ： How do brain mechanisms develop from childhood to adulthood? There is extensive debate if brain development is due to pruning of excess neurons, synapses, and connections, leading to reduction of responses to irrelevant stimuli, or if development is associated with growth of dendritic arbors, synapses, and myelination leading to increased responses and selectivity to relevant stimuli. Our research addresses this central debate using cutting edge multimodal imaging, obtaining multiple measurements of brain function using functional magnetic resonance imaging (fMRI), and brain anatomy using quantitative MRI (qMRI) and diffusion MRI (dMRI) in each of 27 children (ages 5-12) and 30 adults (ages 22-28). We use the face recognition system as a model system to study brain development as it is a well understood cortical system that shows particularly protracted development throughout childhood and adolescence, into adulthood. Both functional and anatomical measurements provide compelling empirical evidence supporting the growth hypothesis. Functionally, results reveal (1) age-related increases in the size of face-selective regions, (2) age-related increases in responsiveness and selectivity to faces, and (3) a developmental increase in neural sensitivity to face identity, which is correlated with an increase in perceptual discriminability of faces. Importantly, this development is specific, occurring in face- but not object- and place-selective regions and cannot be explained by differences in data quality or measurement noise across age groups. Anatomically, we find (1) age-related decreases in T1 relaxation that are associated with increases in macromolecular tissue volume in face- but not place-selective regions, which we validate in histological slices of postmortem brains, (2) this tissue development is correlated with specific increases in functional selectivity to faces, as well as improvements in face recognition, and (3) the largest developmental decreases in both T1 relaxation and mean diffusivity occur close to the gray-white matter boundary of face-selective regions, suggesting that in addition to dendritic complexification increased myelination may contribute to tissue growth. Together, these data suggest a new model by which emergent brain function and behavior during childhood result from cortical tissue growth rather than from pruning.

この発表は顔認知に関して，構造的・機能的に削除される部分，または促進される部分があるかどうかといった発表でした．機能的ネットワークのみを検討している人は多くいますが，構造も検討している論文や発表を聞いたことがありませんでした．NIRSでは構造は見ることはできませんが，私の研究室でもきっと両方検討できると思いました．また，この発表は子どもから大人にかけての縦断的研究を行っており，顔認知に関する確実な脳の発達が検討されていました．
 

発表タイトル       ： EEG attractor landscape in the resting human brain 著者                  ： Takumi Sase セッション名       ： EEG/MEG Modeling and Analysis Abstract            ： 【Introduction】 Neural networks in the human brain spontaneously change their states depending on cognition, perception, and thought. It has been believed that this spontaneous change is reflected in the macroscopic neural phenomenon, namely, electroencephalography (EEG) dynamics. On this point, the EEG microstate has been proposed where the microstate rapidly switches among four classes (Lehmann, 1971). It has been reported that these four classes are associated with diseases (Lehmann, 2005). However, from the dynamical system viewpoint, we can say that the microstate analysis assumes that the underlying dynamics is fixed points because EEG data is directly applied to the clustering analysis. Here, we propose an extended EEG microstate, namely, EEG mesostate x, where x denotes the dimension of tori in the state space and mesostate 0 is equivalent to the conventional EEG microstate. 【Methods】 In total, 80 subjects were participated in the study after giving informed consent. The study was approved by the ethics committee of RIKEN. We recorded EEG signals from 63 electrodes on the scalp during 180 s under an eyes-closed resting condition by a sampling frequency of 1000 Hz. We used (1) the Poincare section analysis and (2) the clustering analysis. Analysis (1) can convert dynamics to ‘statics’, while analysis (2) can separate statics to multistable states. However, in this study, instead of the Poincare section, we used the instantaneous amplitude analysis to preserve temporal information. Furthermore, to observe the attractor landscape visually, we applied a supervised dimensionality reduction method, namely, the linear discriminant analysis (LDA), to 63-dimensional EEG dynamics. 【Results】 We found that mesostate 2 is associated with three metastable states, namely, three 2-dimensional tori. First, because the power spectrum of EEG signals shows a dominant frequency of 10 Hz, corresponding to the alpha wave, we applied the instantaneous amplitude analysis to EEG signals and submitted the converted signals A1(t) to the clustering analysis, where the number of clusters was set to three. Then, we found that three trajectories separately appear in the state space, but there do not exist attractors corresponding to each cluster. Thus, mesostate 1 was denied. Next, because the power spectrum of A1(t) shows a dominant frequency of 0.3 Hz, corresponding to the delta wave, we applied one more the instantaneous amplitude analysis to A1(t) and submitted the converted signals A2(t) to the clustering analysis, where the number of clusters was also set to three. Then, we found that three trajectories also separately appear in the state space and furthermore, we observed three attractors [Please see Figure]. In addition, we validated the abovementioned results by using Kuramoto associative memory model (Aoyagi, 1995).Attractor landscape of resting-state EEG dynamics. The dynamics spontaneously changes among three metastable states. 【Conclusions】 We showed a possibility that three two-dimensional tori underlie the resting human brain. Our finding is reasonable because it is well known that the torus, namely, phase-amplitude cross-frequency coupling phenomenon often appears in EEG dynamics, where the amplitude of fast oscillations is modulated by the phase of slow oscillations. In the future, functional roles of the metastable states identified here should be elucidated.

日本人の理研の方のポスター発表でした．EEGを用いたrestring-state研究であり，自閉症の方ほど，resting-state中の状態変化の回数が多いという発表でした．この発表はLDAやk-meansを使った発表であり，私も知っている解析が行われていてとても分かりやすい発表でした．　メソstateという手法を初めて聞きました．Methodで図がかっこよく，自分の研究にも活かしたいと思いました．
 

発表タイトル       ：　Myelin Water Imaging in Human Brain: Principles, Validation and Applications 著者                  ： OHBM 2017 Local Organizing Committee セッション名       ： LOC Symposia Abstract            ： White matter makes up 40% of brain tissue. Myelin is a critical structural and functional component of white matter that allows rapid and effective information exchange in the brain. Recent animal work shows that myelin is neuroplastic.  Using a rodent model, McKenzie et al. (2014) established the relationship between oligodendrocyte proliferation and learning, showing accelerated oligodendrocyte generation is associated with performance of a complex skill and an absence of motor learning when these cells were genetically blocked. However, much less is known about what changes in myelin are associated with learning or following brain damage in humans. Recently non-invasive imaging techniques have emerged that can characterize myelin in vivo in humans. This symposium will provide suggestions for the implementation of myelin water imaging to index myelin in humans in future work.

この発表はミエリンに関する発表でした．MRIでミエリンまで撮像できること知りませんでした．ミエリンは学部の時の授業で習って以来全く触れておらず，この学会で久しぶりに耳にしました．ミエリンを使った研究はMISLでは行われていないため，研究の理解が難しいと感じました．ミエリンの可塑性に興味を持ちました．
 

発表タイトル       ：　Exploring the neural evidence of mother-infant entrainment: Inter-brain synchronized hemodynamic activity 著者                  ： Yasuyo Minagawa セッション名       ： Brain-to-brain synchrony early in life: What can we learn from different hyperscanning techniques? Abstract            ： Hyperscanning techniques allow the simultaneous recording of brain activity of different subjects. With the advent of sophisticated new tools and techniques over the past decades, it is now possible to study the inter-brain correlations between cerebral activity of a group of interacting subjects as a unique system. Ecologic experimental designs can be adopted to create an interaction between subjects similar to real life social situations, thus, hyperscanning represents a potentially revolutionary new approach, opening new perspectives for understanding the evolution and development of typical and atypical human social interactions. Given these new opportunities, it appears timely and important to reflect and discuss open questions and current challenges and limitations of different hyperscanning techniques. These include (1) review of experimental tasks suited for hyperscanning across different age groups (from infancy to adulthood) and neuroimaging techniques (EEG, NIRS, fMRI); (2) methodological approaches (such as frequency-based connectivity estimators in EEG hyperscanning, and calculation of temporal correlation and Granger-based causality used on hemodynamic data, i.e., obtained with fMRI and NIRS), (3) impact of subjects’ characteristics (such as age and gender) on neural synchrony measures; (4) behavioral correlates of brain-to-brain synchrony. This symposium intends to provide a forum to stimulate the discussion of these and other issues. Clinical implications will be highlighted, particularly with respect to the relevance of early social interaction for mental health across the life-span. In a nutshell, the symposium aims at providing up-to-date knowledge on hyperscanning techniques of social interactions during human development. Each presenter brings long-standing unique and complementary expertise to the table, making the sum greater than the parts.

この発表はハイパースキャニングに関する発表で，赤ちゃんと母親の同時計測研究でした．この研究で私が注目したのはNMFという解析手法とウェーブレット変換による前処理です．現在，前処理ではバンドバスがMISLでは一般的ですが，ウェーブレットによる前処理とどちらが良いのか検討する必要があると感じました．また，前頭部だけ測定をしているのでNMFも使えるかもしれないと考えました．NMFはほとんど詳しくないので，調査をしようと思いました．
 

発表タイトル       ：　Meditation-Inspired Cognitive Training Improves Working Memory and Increases Cortical Thickness 著者                  ： David Ziegler セッション名       ： ポスターセッション Abstract            ： 【Introduction】 Attention can be oriented externally to the environment or internally to the mind, and can be derailed by interference from irrelevant information originating from either external (e.g., distracting sights or sounds) or internal sources (e.g., distracting or intrusive thoughts). We designed a mobile meditation-inspired training app (MediTrain) that draws from focused-attention meditation practices. MediTrain is a tablet-based, meditation-inspired cognitive training game aimed at improving self-regulation of internal distraction. It was developed in collaboration with Jack Kornfield, a meditation thought leader, and Zynga, a world-class video game company. This game is designed to make benefits of meditation easily accessible to anyone, including complete novices. We achieved this by creating a game experience that yields quantifiable and attainable goals, provides both punctuated and continuous feedback, and includes an adaptive algorithm to increase difficultly as users improve. We hypothesized that MediTrain would improve participants’ attention, supporting the ability to maintain information in working memory (WM) while regulating internal distractions and avoiding external distractors.   【Methods】 Before and after 6-weeks of training with MediTrain (n = 24, 13F) or an active placebo training app (n = 20, 12F), healthy young adults performed an attention-demanding task requiring high and low load visual WM and distractor filtering (the Filter Task), a test of working memory capacity (Change Localization Task) and underwent structural MRI. Both training programs were completed on a mobile platform (iPad), 5 days per week, with an average training time of 25 minutes per day. To test for training effects, we performed an ANCOVA on Filter Task accuracy at post-testing with pre-testing included as a covariate.   【Results】 We found that participants who completed training with MediTrain showed significantly improved accuracy compared with placebo on both Filter Task conditions with low WM load (Set 1, distractors present, p = 0.02; Set 1, no distractors present, p = 0.02). MediTrain participants also showed improvements in accuracy compared with placebo during conditions with high-load WM (Set 3) with no distractors present (p = 0.03). For the Change Localization Task, a “K score” is calculated for each participant, providing an index of their overall working memory capacity before and after training. After six weeks of training, the MediTrain group is showed a significant increase in K, while the placebo group remained unchanged (p = 0.04). Structural MRI data were processed with the semi-automated Freesurfer anatomical pipeline for longitudinal anatomical data and surface-wise GLMs were performed to compare differences in the rate of change in cortical thickness between training groups. After Monte Carlo correction for multiple comparisons, three clusters showed a significant increase in cortical thickness in the MediTrain group, compared to controls. These clusters were located in the medial orbitofrontal cortex, lateral prefrontal cortex, and superior temporal gyrus.   【Conclusions】 These results show that 6-weeks of training with our novel approach to meditation leads to increased working memory abilities and is also associated with increased cortical thickness in areas associated with cognitive control, self-regulation, and interoception.

この発表は瞑想を用いたトレーニングアプリを用いて可塑性を検討した発表でした．瞑想を数週間続けることによる皮質の厚み増加の論文は見たこと有りましたが，この発表は簡易に瞑想ができるアプリを使って検討をしていました．これは私が目指している簡易な装置によるフィードバックと深く関わりがあり，このアプリがどのような仕組みになっているのか気になりました．
 
参考文献

• 23^rd Annual Meeting of the Organization for Human Brain Mapping, https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageID=3734

学会参加報告書

報告者氏名	中村圭佑
発表論文タイトル	SLICとNormalized Cutを用いた脳領域の分割手法
発表論文英タイトル	Brain region segmentation method using SLIC and Normalized Cut
著者	中村圭佑，日和悟，廣安知之
主催	Organization for human brain mapping
講演会名	OHBM 2017
会場	Vancouver convention centre
開催日程	2017/6/25-2017/6/29

 
 

1. 講演会の詳細

2017/6/25-29に，カナダのバンクーバーにて開催されましたOHBM 2017に参加いたしました．OHBM2017 は，Organization for human brain mappingによって主催された国際会議で，ヒトの脳組織に関する研究に携わる様々な背景の研究者を集め，これらの科学者のコミュニケーション，および教育を促進することを目的に開催されています．本研究室からは他に廣安先生，日和先生，池田さん，石田翔也くん，片山さん，石原さん，萩原さん，玉城さん，吉武さん，相本くん，藤井聖香さん，三好くん，水野さんが参加しました．
 

2. 研究発表
   • 発表概要

私は26日のポスターセッションに参加いたしました．発表の形式はポスター発表で，2時間自由に参加者の方と議論を行いました．
今回の発表は，「Brain region segmentation method using SLIC and Normalized Cut」と題して発表いたしました．以下に抄録を記載します．

Introduction：In the brain function analysis, identification of ROI (Region Of Interest) is critical in considering the activation site and functional connectivity of the brain. The shape and location of the ROI in the brain is defined by the brain atlas. However, most of the brain atlas are those labeled based on the brain anatomical division of the brain part, or regions labeled with functional connections. Therefore, in the former brain atlas, the number and size of ROIs are arbitrarily changed, and it is hard to examine the brain function. Also, since the shape of the brain varies among individuals, it is difficult to consider the difference in brain shape in the method of determining the ROI using the conventional atlas. In this study, we developed an automatic brain atlas creation system that can change the number of region divisions. Methods：In the proposed method, structural images and functional images by MRI (Magnetic Resonance Imaging) are segmented by SLIC (Simple Linear Iterative Clustering) and Normalized Cut [. In this method, the number of divisions of the brain region is changed by changing parameters at the time of image division. To confirm the effectiveness of the method, the proposed method was applied to the T1 image on which the standard cerebralization process was performed. The T1 images of two men were used, and the division of the obtained brain regions was compared. For quantification of regional division, the similarity was calculated by Jaccard Index. Results：By changing the parameters of SLIC and Normalized Cut, about 150-300 regions were divided. The results were good division results along the anatomical division.The average value of similarity of the region shape of the two subjects was about 23% when divided into 210 regions. At this time, the similarity of the region shape was 56% at the maximum and 3% at the minimum. Therefore, the results of region segmentation of this method showed that differences occur depending on the position of the brain region. Conclusions：In this study, a brain atlas creation system based on brain region division by arbitrary division number using SLIC and Normalized Cut was proposed. The proposed method was applied to the T1 image of two subjects, and the division result by the Jaccard Index was quantified. As a result of region segmentation by the proposed method, the T1 image was divided into 150 to 350 regions. According to the Jaccard Index of the region shape after the division of the two subjects, regions with high similarity and regions with low similarity existed.

 

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
 
・質問内容1
複数名の方からSLICの分割で用いている初期クラスタのKの値はいくらかという質問をいただきました．こちらの質問に対する回答ですが，今回のケースでは1024領域に分割しているが，もしより細かい分割が必要であるならば，Kの値を大きくする必要がある，と回答いたしました．
 
・質問内容2
Fiber Trackingの際に用いているパラメータについての質問をいただきました．Fiber Trackingを行う際のパラメータを控えていなかったため，その旨を返答致しました．今後のこのようなことが無いよう，使用したツール・機器の設定/パラメータを控えておくように心がけます．
 
・質問内容3
評価に用いているDice係数の値が低いが，個人のデータごとに分割を行う際にどのような要因によりクラスタリング結果に違いが生じているのかという質問をいただきました．この質問に対し，Fiber Trackingの結果が個人により異なるため，生成されるアトラスも異なる分割となっている．と答えました．
 

• 感想

今回は初めての学会発表であり，また初の英語発表ということで，自身の発表をどこまで伝えられるか，自身の研究のどこをアピールできるのか，ということを考えて準備してきました．今回のポスター発表では，ポスターのファーストインプレッションという点ではセッションの最初であまり人の目を引くことが出来ず，ポスターの見た目のインパクトと題目の重要さを身をもって痛感し，研究のアピールという点では不十分であったと思います．この点は，今後の発表で活かしたいと思います．一方，セッション終盤では自分と同じParcellationの分野に携わっている方々と意見を交わすことができ，自分の研究を十分に伝え，かつ有意義な議論ができたと感じました．また，その他のParcellationの研究やAtlas basedアプローチを行っている研究について聴講し，脳機能研究における自分の研究の立ち位置を把握することが出来，今後の自分の研究に励みになったと考えています．
 

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました．
 

発表タイトル      ：　Performance of Various Brain Atlases for Individual Idetification using resting fMRI 著者              ： Andrew Michael, Chao Zhang セッション名       ： Oral Session(Informatics) Abstruct        ： Introduction: In a recent paper (Finn et al. 2015) it was demonstrated that functional connectivity (FC) features based on resting fMRI (rfMRI) data can be used as a “fingerprint” to accurately identify individual subjects. Subsequent studies have since demonstrated the utility of rfMRI in predicting individual differences in a wide array of cognitive measures such as intelligence (Hearne, Mattingley, and Cocchi 2016), distractibility (Poole et al. 2016), and attention (Rosenberg et al. 2015). The finding that rfMRI can be utilized to predict individual differences has paved the way for the potential use of functional “fingerprints” in the treatment and diagnosis of psychiatric disorders. However, the above studies were mostly based on a small number of subject (N~100). In this study we address the following questions: (1) can rfMRI FC predict individuals in a large cohort? (2) what brain atlases are the best performers for individual identification? (3) can rfMRI data acquired on a different day be used for individual prediction? and (4) what is the separation between the best identification and the next best match? Methods: Our study contained rfMRI from 820 healthy young adults (366 males and 454 females, age: 22-37 years) from the Human Connectome Project (HCP) S900 release (Van Essen et al. 2012). Each subject was scanned on four different runs (2 each on 2 separate days). For each run and each subject, time series information was extracted from ROIs as defined by the following ten different brain atlases: DOS160 (Dosenbach et al. 2010); CC400/CC200 (Craddock et al. 2012); AICHA (Joliot et al. 2015); Stanford90 (Shirer et al. 2012); Harvard-Oxford; Automated Anatomical Labeling (AAL); AAL_new; AAL2; and Brodmann. Individual FC was calculated between ROI time series using Pearson correlation. To implement individual identification, we correlate the FC of each subject from Run1 to FC of all 820 subjects from Run2 and paired subjects based on maximum correlation. The prediction accuracy is defined as the proportion of subjects with correct identification. We then repeat this process to identify subjects using Run3 and Run4 data. Results: Prediction accuracies for the ten different atlases are presented in Figure 1. DOS160 produced the highest accuracy of 95%. The accuracies of four other functional atlases were above 80%. We note that the performance of the five structural atlases was in the range of 54–66%. Prediction accuracies for Run3 and Run4 data were 95% and 88% respectively (Figure 2). In Figure 2 we further investigate the FC correlation of the correctly identified subjects (in red) and the next best 20 FC correlations (in blue). We note that for a large proportion of the 820 subjects, the second best match is significantly lower than the correct match indicating the robustness of rfMRI FC for individual identification. Conclusions: We performed individual identification using rfMRI data for a large cohort of 820 subjects and show that the DOS160 atlas is the best performer. Of the atlases examined, the five functional parcellations demonstrate much higher identification accuracies (above 80%) than the five structural parcellations (<66%). We show that high prediction accuracies are possible between rfMRI data acquired on different days. We conclude that choice of parcellation scheme is an important consideration for studies performing individual identification. By improving characterization of FC differences at the individual level, it may be possible to gain novel insights into the association between individual FC differences and distinct cognitive or behavioral features.

本発表は，脳の機能的コネクティビティを用い，”figer print”として個人を同定することを目的とするものでした．発表においては，用いるアトラスによって同定の精度が異なるという結果が示されており，機能的コネクティビティを考慮する際におけるアトラスの重要性に気づかされました．
 

発表タイトル      ： Brainnetome Atlas: A New Map of Human Brain 著者              ： Lingzhong Fan, Hai Li, Zhengyi Yang, Tianzi Jiang セッション名       ： Oral Session(Informatics) Abstruct        ： Introduction: The brain atlases based on different mapping techniques are the navigator of the human brain, and considered as the cornerstone of basic and clinic neuroscience(Toga et al., 2006; Evans et al., 2012; Amunts & Zilles, 2015). With a history of more than a century, the Brodmann’s map developed by a neuroanatomist, Korbinian Brodmann, divided the human cerebral cortex into 52 different areas based on its cellular architecture, is still used most often as one of the possible parcellations(Zilles & Amunts, 2010). However, the limitations of this map have become more and more obvious, increasing the importance of defining brain areas using new methodologies(Paxinos, 2016). In the year of 2010, the Brainnetome project was launched to investigate the hierarchy in human brain from genetics and neuronal circuits to behaviors. One of the key elements of this project is focused on setting up and optimizing the framework for connectivity-based parcellation, and aims to produce a new human brain atlas, i.e. Brainnetome atlas based on connectional architecture (Jiang, 2013; Fan et al., 2016).Currently, the human Brainnetome Atlas is freely available for download at http://atlas.brainnetome.org, so that whole brain parcellations, connections, and functional data will be readily available for researchers to use in their investigations into healthy and pathological states. Methods: Using noninvasive multimodal neuroimaging techniques, we designed a connectivity-based parcellation framework that identifies the subdivisions of the entire human brain, revealing the in vivo connectivity architecture. The resulting human Brainnetome Atlas, with 210 cortical and 36 subcortical subregions, provides a fine-grained, cross-validated atlas and contains information on both anatomical and functional connections. Additionally, we further mapped the delineated structures to mental processes by reference to the BrainMap database. As part of this work, we developed an integrated “Automatic Tractography-based Parcellation Pipeline (ATPP)” to realize the parcellation using automatic processing and massive parallel computing (Fig. 1) that we share with the atlas. Results: This new brain atlas has the following four features(Fig.2): (A) It establishes a fine-grained brain parcellation scheme for 210 cortical and 36 subcortical regions with a coherent pattern of anatomical connections; (B) It supplies a detailed map of anatomical and functional connections; (C) it decodes brain functions using a meta-analytical approach; and (D) It is an open resource for researchers to use for the analysis of whole brain parcellations, connections, and functions.The Brainnetome Atlas together with its related software is available for download to serve as a shared community resource. The pipeline software is open to the community to facilitate the parcellation of specific brain regions of interest. Conclusions: The human Brainnetome Atlas could constitute a major breakthrough in the study of human brain atlas and provides the basis for new lines of inquiry about the brain organization. It will enable the generation of future brain atlases that are more finely, defined and that will advance from single anatomical descriptions to an integrated atlas that includes structure, function, and connectivity, along with other potential sources of information. It will present neuroscientists with one of the key tools that will help us get some entirely new knowledge on how the brain works, as well as to understand the pathophysiological mechanism of psychiatric and neurological disorders.

 
本発表は，機能的コネクティビティ，および神経追跡を用いて脳アトラスを作成する手法についての発表でした．上記のデータを用いてデータドリブンなアプローチでアトラスを作成する手法であり，2016年に発表されたものでしたが，OHBM中にこのアトラスを用いた研究を複数みかけました．自身のアプローチに近い手法であるため，今後どのようにアトラスを作っていくかを考えさせられました．
 

発表タイトル      ：Adaptive Cortical Parcellaions for Source Reconstructed EEG/MEG Connectomes 著者              ：Seyedehrezvan Farahibozorg, Richard Henson, Olaf Hauk セッション名       ： Oral Session(Modeling & Analysis) Abstruct        ： Introduction: There is growing interest in the rich temporal and spectral properties of Electro- and Magnetoencephalography (E/MEG) signals in order to study the functional connectome of the brain [1, 2]. However, the spatial resolution of E/MEG data is limited, because several thousand sources of activation in the brain must be estimated from maximally a few hundred recording sites. This limited spatial resolution causes the so-called leakage problem: activity estimated in one region of interest (ROI) can be affected by leakage from locations outside this ROI [3, 4]. E/MEG studies typically adopt parcellations from structural or fMRI research for whole-brain connectivity analysis [5]. However, considering the spatial resolution of E/MEG, these parcellations are unlikely to be optimal [6]. Here, we utilise Cross-Talk Functions (CTFs) as a direct measure of spatial leakage [7] and utilise two CTF-informed image segmentation algorithms in order to parcellate the cortical surface into the maximum number of distinguishable ROIs. Methods: We computed resolution matrices (with rows as CTFs) for individual subjects, based on forward and inverse models computed using BEM head models and L2 MNE inverse operators of 17 healthy subjects. In the first parcellation approach, we started from standard anatomical parcellations and modified the ROIs using a CTF-informed split-and-merge (SaM) algorithm [8]. In the second approach, we started from all brain vertices with no prior parcellation. A CTF-informed region growing (RG) algorithm [8] was used to create ROIs around the vertices that showed highest sensitivity and specificity of CTFs on the cortex, which were then optimised using an SaM algorithm. The algorithms are designed such that they merge ROIs/vertices with highly overlapping CTFs, split ROIs that produce distinguishable patterns of CTFs, remove ROIs with low sensitivity, and for each ROI identify a group of representative vertices that show high sensitivity and specificity to that particular ROI. We used ROI Resolution Matrices (RRmat) to quantify leakage from each ROI to all other ROIs in the brain in order to evaluate the parcellations’ performance where an ideal RRmat is an identity matrix. Thereafter, we evaluated the possible consequences of using different parcellation methods for graph-theoretical connectivity analyses on simulated data with realistic levels of noise. Results: Based on the RRmats (Fig. 1), we found that parcellation sensitivity improved from 0.47 and 0.37 in two standard anatomical parcellations (Desikan-Killiany (DKA) and Destrieux Atlases (DA) respectively) to 0.65, 0.70 and 0.70 in modified DKA, DA and RG parcellations respectively. Moreover, ROI distinguishability improved from 0.50 and 0.38 to 0.61, 0.65 and 0.64 (Fig. 1). Interestingly, in spite of their different starting points, both SaM and RG algorithms yielded approximately 70 ROIs. Furthermore, our simulated realistic connectome with a single hub showed that modified parcellations were particularly successful in improving hub sensitivity and hub connectivity probability patterns (Fig. 2). Conclusions: Our proposed parcellation algorithms significantly improved the sensitivity and distinguishability of ROIs compared to the anatomical parcellations, while at the same time maximising the number of distinguishable ROIs in the brain. The algorithms are adaptive with respect to the measurement configuration and source localisation methods. Regardless of the starting point they yielded around 70 ROIs, suggesting that this reflects the resolution limit of this particular sensor configuration and source estimation method. Furthermore, our simulations showed that the choice of parcellation can have significant impact on the outcome of graph theoretical analysis of the source-reconstructed E/MEG. Therefore, we conclude that adaptive parcellations are essential for whole-brain EEG/MEG connectomics.

 
本発表は，EEGおよびMEGを用いて機能的コネクティビティを考慮する際に発生するleakage problemに対して，用いるアトラスの形状とサイズをCross-Talk FunctionによってE/MEGにとって最適なものにすることで，leakage problemの解決を試みる発表でした．NIRSについても同様のことが出来そうな内容であり，考慮する必要性を感じました．
 

発表タイトル      ：　Unravelling the intrinsic functional boundaries of the macaque monkey cortex 著者              ： Ting Xu, Alexander Opitz, Arnaud Falchier, Gary Linn, Deborah Ross, Julian Ramirez, Darrick Sturgeon, Eric Feczko, Elinor Sullivan, Jennifer Bagley, Stan Colcombe, Damien Fair, Charles Schroeder, Michael Milham セッション名       ： Oral Session(Modeling & Analysis) Abstruct        ： Introduction: A growing body of literature has demonstrated the ability to delineate cortical areas in the human brain based upon the detection of spatial transitions in intrinsic functional connectivity (iFC) profiles (Cohen et al., 2008; Wig et al., 2014). In particular, gradient-based parcellation approaches have gained popularity due to their ability to recapitulate previously established cytoarchitectonic brain areas. Here, we demonstrate the feasibility of extending the application of parcellation approaches to non-human primates (NHP), demonstrating the reliability of these parcellations and comparing the cortical areas revealed to those obtained in humans. Methods: We collected data from a male rhesus macaque monkey (age: 6 year) on a 3 Tesla Siemen Tim Trio scanner. Awake functional MRI scans were obtained during 6 sessions (4-7 scans for each session, 8 minutes per scan, 216 minutes in total, TR = 2 s, 1.46 x 1.46 x 2 mm); three of the sessions were carried out using a contrast agent (i.e., monocrystalline iron oxide particle (MION)) and 3 were without contrast. We obtained high-resolution T1-weighted anatomical images (0.5mm isotropic voxel) for surface registration. The native surface was reconstructed and registered to Yerkes19 macaque template (Donahue et al., 2016). We calculated iFC-similarity maps for each scan, followed by the spatial gradient and edge detection computation on native surface. The spatial correlations were calculated to investigate the reproducibility of boundaries across sessions and scans. We further explored the requirement of scan time for a relatively robust iFC and boundary map. Results: As expected, whole-brain gradient maps exhibited a higher degree of similarity among individuals within the same developmental period; differences were particularly notable at the extremes (i.e., childhood, older age) (see Figure 1A). To facilitate visualization, we defined 6 age groups and depicted mean gradient maps in Figure 1B. Next, at each voxel, we used univariate analyses to detect age-related linear and quadratic trends in global mean for the gradient map associated with that specific vertex. These analyses revealed linear age effects in posterior cortex, particularly in primary visual, sensorimotor, and default mode networks (Figure 1C). The quadratic effects were mainly located in the regions of network borders e.g. default mode, ventral attention (Figure 1C). Finally, at each vertex, we used MDMR to detect age-related variation (linear, quadratic) in the gradient maps defined across individuals. The linear and quadratic age-related effects were predominantly located in the regions of network borders, e.g. default mode, ventral attention, dorsal attention and frontoparietal network (Figure 2). Conclusions: By examining the transition pattern of iFC similarity in macaque, we have demonstrated the ability to detect functional boundaries and cortical areas in the macaque monkey cortex using awake R-fMRI in macaque, suggesting a reliable scheme for delineating cortical organization in macaque and potential utility for validating invasive individual parcellation.

 
本発表は，マカクザルの脳の機能的境界をアトラスを用いて定義し，そのアトラスを用いてマカクザルの脳機能を解析する，といった内容でした．アトラス作成に用いられたgradient-based parcellationと評価の方法については，今後の研究で機能データを用いる際の参考になると思いました．
 

発表タイトル      ： Predicting Personality from Network-based Resting-State Functional Connectivity 著者              ： Alessandra Nostro, Veronika Müller, Deepthi Varikuti, Rachel Pläschke, Robert Langner, Simon Eickhoff セッション名       ： Oral Session(Social Neuroscience) Abstruct        ： Introduction: Personality as a key feature of inter-individual differences affects all aspects of life, including affective, social, executive and memory functioning [3,4,6]. Task-based fMRI studies investigated personality and brain activity in association to each of these domains; however, since personality traits are enduring across situations [2], it is possible that they relate to many brain systems, not detected by task-based fMRI. The investigation of functional connectivity in resting state conditions might therefore help in capturing the intrinsic and complex neural architecture underlying personality [1]. A recent study [7] showed a sexual dimorphism in brain structure-personality relationships, with associations revealed only in males. In females, brain connectivity rather than structure, might thus play a stronger role in light of personality. Therefore, we aimed to predict scores of the five-factor personality model (openness, conscientiousness, extraversion, agreeableness, neuroticism) [2] from resting-state functional connectivity (RS-FC) in meta-analytically defined brain networks, and tested how these predictions are modulated by gender. Methods: We assessed 9 meta-analytic networks representing regions consistently activated by different social (empathy, face perception), affective (reward, pain, emotion perception), executive (working memory, vigilant attention) and mnemonic (autobiographic and semantic memory) functions. FIX-denoised RS fMRI data of 136 males and 137 matched females was downloaded from the HCP WU-Minn Consortium [10] and further preprocessed with SPM8 using standard procedures. Within each network, FC between all nodes was computed using their respective extracted time series. A relevance vector machine-learning algorithm [9] was used to predict NEO-FFI scores [2] based on FC between all nodes of each network, separately for males and females. Prediction performance was assessed by Pearson correlations between real and predicted scores (p<0.05, corrected for multiple comparisons) and compared between groups. Results: Personality traits were successfully predicted by FC within different networks in men and women (see Fig. 1 for a summary). Specifically, in men, conscientiousness was predicted by FC within networks of the affective system (e.g. r=.40 for the reward network; Fig. 2A), extraversion by networks related to social, memory and affective processing, and agreeableness by networks of affective and social domains. In women, openness was predicted by FC within affective and memory-related networks (e.g. r=.45 for the autobiographic memory network; Fig. 2B), conscientiousness by networks linked to executive functioning, and neuroticism by memory-related network. Significant gender differences in prediction performance were found for openness, conscientiousness and agreeableness (Fig. 1). Conclusions: Using machine-learning techniques the current study revealed substantial associations of personality with various brain networks related to affective, social, executive, and long-term memory functions, based on FC within these networks. These results indicate that RS connectivity patterns within meta-analytically defined functional brain systems provide information on the individual expression of specific personality traits. Indeed, they were not only predicted by networks already associated to them in the literature, but also not expected brain systems were found informative, with the exception of neuroticism which was not predicted by any expected affective networks. Additionally, FC patterns of different functional networks were shown to predict different personality traits in males and females, indicating gender-specific neural mechanisms associated with specific personality characteristics. This extends previous findings on relations between network-specific differences in gray-matter volume and personality [7] by demonstrating that RS-FC–personality relations should not be considered independent of gender.

 
本発表は，resting stateにおける脳の機能的コネクティビティを解析することで，個人の性格を推定する，といった内容でした．性格の各要素に関連する脳の機能的ネットワークをmeta analysisを用いて抽出しており，今後の研究においてこのような手法を活用できるのではないかと思いました．
 
参考文献

• OHBM 2017,

https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageID=3734
 
学会参加報告書

報告者氏名	玉城貴也
発表論文タイトル	Human state estimation from cerebral blood flow data using convolutional neural network and long short-term memory
発表論文英タイトル	Human state estimation from cerebral blood flow data using convolutional neural network and long short-term memory
著者	玉城貴也，日和悟，蜂須賀啓介，奥野英一，廣安知之
主催	Organization for Human Brain Mapping
講演会名	The 23nd Annual Meeting of the Organization for Human Brain Mapping
会場	Vancouver Convention Centre
開催日程	2017/06/25-2017/06/29

 
 

1. 講演会の詳細

2017/06/25から2016/06/29にかけて，カナダのバンクーバ（Convention Centre）にて開催されましたThe 23nd Annual Meeting of the Organization for Human Brain Mappingに参加いたしました．この学会は，ヒト脳の高次機能を様々なイメージング装置によって解明するために，最新かつ革新的な研究の情報を交換することや研究成果について議論することを目的に毎年開催されています．私は25日から29日の全日に参加いたしました．本研究室からは他に廣安先生，日和先生，石原，和田，萩原，吉武，片山，相本，石田翔也，三好，中村圭介，池田，水野，藤井が参加しました．

2. 研究発表
   • 発表概要

私は28日の午後のPoster Sessionに参加いたしました．発表の形式はポスター発表で，2時間自由に参加者の方と議論を行いました．
今回の発表は，「Human state estimation from cerebral blood flow data
using convolutional neural network and long short-term memory」　という題目で，Deep Learningによるヒトの内部状態の推定と機械学習からの重要領域に関する知識獲得について発表を行いました．以下に抄録を記載致します．
 

【Introduction】 With the development of machine learning technology in recent years, the estimation of human internal state from brain activity has attracted considerable attention. A drawback of the conventional method is that pre-processing of the feature extraction is required before learning, which necessitates prior knowledge about the learning data. To solve this issue, in this study, we investigated human internal state estimation using a deep neural network. Furthermore, as the constructed learner expresses the relationship between brain activity and human state, a method to obtain knowledge about the important brain regions and time segments from then was studied. 【Methods】 As human brain activity has spatial dependence and time-varying characteristics between different brain regions, a machine learning algorithm that can deal with these is necessary. We proposed a new algorithm that is integrated with a convolutional neural network [1] expressing spatial dependence and long short-term memory (LSTM) [2], which deals with time-varying characteristics. Our proposed learner comprised five layers: input, convolution, LSTM, neural network, and softmax, as shown in Fig1. The learner was trained to classify the brain activity data of the forehead in the N-back task (N = 2, 3) of 10 healthy men measured by functional near-infrared spectroscopy as either 2-back or 3-back. Furthermore, as a method of extracting important brain regions from the constructed classifier, we analyzed the sensitivity of the measurement channel. Sensitivity was calculated from the variation of the error function when the average of brain activity data of all subjects for each measurement channel as input data, adding the standard deviation of the data of each measurement channel as variation. 【Results】 Our learner achieved an accuracy of 98.00 ± 3.50%, and it was shown that classifying the brain state during the N-back task with high accuracy is possible. The sensitivities of 2-back and 3-back mean brain activity data were compared. The sensitivity was nearly zero in every measurement channel in 3-back, whereas in 2-back, high sensitivity in specific brain regions was observed. This result suggests that the learner expressed the features of brain activity mainly at the 2-back task and classified the two states. Moreover, the brain regions with high sensitivity in the 2-back task were dorsolateral prefrontal cortex (DLPFC), anterior prefrontal cortex (APFC), ventrolateral prefrontal cortex (VLPFC), and orbitofrontal cortex (OFC). DLPFC and APFC have been reported to be active in numerous working memory tasks and are involved in the retention of information in spatial working memory [3]. VLPFC have been reported to be active in a verbal working memory task [4]. OFC is activated when it is required to redirect attention from the failure of the task [5]. These observations suggest that extracted brain regions are reasonable. 【Conclusions】 We developed a novel machine learning algorithm to estimate the state of a human without special knowledge and showed that the learner could classify the task load from the cerebral blood flow data during a working memory task. Moreover, the brain regions related to the task were extracted by sensitivity analysis of the input channel of the classifier. In conclusion, it was shown that the proposed method is useful as a means of estimating human state.

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
 
・質問内容1
National Institute of Mental Health所属のJong-Hwan Leeさんからの質問です．こちらの質問はなぜDNNの識別率が50%（チャンスレベル）程度なのかというものでした．この質問に対する私の回答は提案手法が空間的特徴量を抽出するConvolution部分のフィルタを時系列間で共有しているため，学習するパラメータが少ないのに対して，DNNは全ての時系列の全チャンネルのfNIRSデータを重み結合で次元削減する必要があるため，学習パラメータが多く，このような結果となったと回答しました．
 
・質問内容2
質問者の氏名を控え損ねてしまいました．こちらの質問はなぜ感度解析の際に被験者データを平均化する必要があるのかというものでした．この質問に対する私の回答は，感度解析は学習済みの識別器から重要な入力チャンネルを抽出することを目的としており，そのため，識別器の平均特性として，その標準を変量としたときにどのチャンネルが感度が高いかをみることで， 重要な入力チャンネルが特定できると回答しました．
 
・質問内容3
質問者の氏名を控え損ねてしまいました．こちらの質問は比較手法と比べて，何が決定的に違うのかというものでした．この質問に対する私の回答は，比較手法と決定的に異なるのは，入力の脳活動データの空間的特徴量と，時間特徴量を分けて抽出する点であると回答しました．
 
 
 

• 感想

前回の国際発表と比べて聴講者の数も多く，反応も良かったので自分の成長を感じることができました．そして，Deep Learningのオーラル発表と自分の研究を比較しても大きな差はないと感じたため，今後も精進して研究に励み，他の研究者に負けないような結果を出そうと思いました．

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました．

発表タイトル       ：　Deep neural network predicts emotional responses using whole brain neuronal activations 著者                  ： Hyun-Chul Kim セッション名           ： ORAL SESSION: Emotion and Motivation Abstruct ： 【Introduction】 Deep neural network (DNN) with an explicit sparsity regularization scheme has been proven useful for functional magnetic resonance imaging (fMRI) data to address the curse-of-dimensionality issue [4, 6]. In this study, we were motivated to investigate the utility of the DNN to regress (DNNR) emotional responses and concurrently to extract the emotional circuity using the whole brain neuronal activations. 【Methods】 Healthy right-handed males (n = 10; 27.2 ± 6.9 years) listened to each of 80 International Affective Digital Sounds [1] across four fMRI runs (20 trials per run) and rated their emotional scores via the nine-scale self-assessment manikins [1] (Fig. 1(a)): arousal (1: very calm, 9: very aroused), dominance (1: very dominated, 9: very dominant), valence (1: very unhappy, 9: very happy).FMRI data were acquired using the standard echo-planar-imaging (EPI) pulse sequence (repetition time/echo time = 2000/30ms, flip angle = 90◦, voxel size = 3.75 × 3.75 × 4mm3, 36 axial slices with no gap). Raw EPI volumes were preprocessed using the SPM8 and ArtRepair [7] toolboxes. Using the preprocessed EPI data, neuronal activations evoked from each of all affective sounds were estimated via the general linear model. Neuronal activations (i.e. 55,417 voxels within the whole brain) were used as an input of the DNNR and support vector machine based regressor (SVMR) to predict participant’s emotional scores in the output.The DNNR with three hidden layers were used as shown Fig. 1(a). A hyperbolic tangent was used as a hidden node activation function and a linear function was used at an output node. The Hoyer’s sparseness (HSP) [3, 5] and DNN node-wise control of weight sparsity were applied in comparison to our earlier studies [4, 6]. Using the nested five-fold cross-validation (CV) scheme (Fig. 1(b)), all 27 combinatorial scenarios of three target HSP levels (i.e. 0.3, 0.5, and 0.7) across three hidden layers were systematically validated to find optimal HSP levels from the training and validation data. The emotional score prediction was then evaluated using the test data. The Python based DNN toolbox (github.com/lisa-lab/DeepLearningTutorials) was modified to implement our explicit L1 norm regularization scheme. The linear combination of weight matrices across hidden layers was obtained to interpret the trained DNNR [4-6]. The linear- and non-linear-kernel SVMRs [2, 8] in the LIBSVM toolbox [2] were also used to compare the performance from the DNNR. Hyper parameters of the SVMRs were systematically validated and the predicted emotional scores were evaluated in the nested five-fold CV scheme. 【Results】 As shown in Fig. 2, Pearson’s correlation coefficients between participants’ emotional scores and predicted scores from the DNNR (mean ± standard error; 0.45 ± 0.07, 0.47 ± 0.08, and 0.48 ± 0.02 for arousal, dominance, and valence, respectively) were significantly greater (Bonferroni-corrected p-value < 10-3) than the SVMR with the linear-kernel (0.18 ± 0.13, 0.12 ± 0.08, and 0.07 ± 0.11) and the non-linear-kernel (0.11 ± 0.14, 0.02 ± 0.12, and 0.05 ± 0.13). Overall, the DNNR features showed the strong negative intensities in the auditory areas, whereas these were mixture of positive/negative intensities such as in the anterior cingulate cortex, insula, and orbitofrontal cortex. 【Conclusions】 The performance of the DNNR was superior to the SVMRs in both (a) the automatic extraction of distinct features associated with human emotional processing and (b) the prediction of emotion scores based on the combination of these distinct features at the output layer.

この発表はDeep Learningによる感情スコアの予測と識別器からの重要入力データの抽出についての発表でした．　線形乗算により重要領域を求めていましたが，DNNのような非線形識別にも線形乗算が有効であるのか疑問でした．
 

発表タイトル       ：Deep Recurrent Neural Network Reveals A Hierarchy of Temporal Receptive Window in the Visual Cortex 著者                  ： Junxing Shi セッション名       ： ORAL SESSION: Perception & Attention Abstruct            ： 【Introduction】 How does the brain support natural vision? Recent studies have shown that Convolutional Neural Networks (CNNs) uncover a representational hierarchy from the striate to the extra-striate cortex that matches a spectrum of unfolded spatial features [2,3,5,6]. However, CNNs, as models of the visual system, have a fundamental drawback in that they only consider instantaneous frames and disregard the temporal structure embedded in dynamic visual inputs. In contrast, the brain integrates information not only in space, but also in time. The time window within which the past information affects the current response is known as the Temporal Receptive Window (TRW) [4]. Therefore, we integrated temporal structures into a CNN and demonstrated for the first time that the model reveals a hierarchy of TRWs along cortical pathways. 【Methods】 We acquired the BOLD-fMRI response from subjects watching a collection of natural movies, with their eyes fixated at the center of the screen. The collection of movies consisted of training (10 hrs), validation (24 min), and testing (8 min) parts. We constructed an encoding model to predict the response at each cortical location by approximating its function using a nonlinear visual model and a linear projection model (Figure 1A). The visual model, integrated with a pre-trained CNN, contained four layers of a Convolutional Gated Recurrent Unit (CGRU) and was trained for an action recognition task with videos [1]. A CGRU takes its previous state through a gating mechanism and the current frame through the CNN as inputs, and outputs the current state as the data representation. The gating mechanism determines how much of the previous state is passed into the current state. As such, the visual model captures the temporal dynamics of the visual stimuli and learns hierarchical spatio-temporal features from videos. We extracted the hierarchical representations by feeding the visual model the same collection of movies as was presented to the subjects.The training and validation movies were used to train, by 5-fold cross-validation, the L2-regularized linear projection models. The models mapped the extracted representations to the BOLD-fMRI responses at individual cortical locations using a homogeneous hemodynamic response function (HRF). Finally, we used the testing movie to perform univariate correlation analysis, evaluate the performance of the projection models, and calculate the TRWs at individual cortical locations. To determine the TRWs at individual cortical locations, we utilized two facts: 1) every unit in the visual model is associated with its own TRW, and 2) the linear projection model associates every cortical location with a set of units in the visual model. 【Results】 As shown in Fig. 2, Pearson’s correlation coefficients between participants’ emotional scores and predicted scores from the DNNR (mean ± standard error; 0.45 ± 0.07, 0.47 ± 0.08, and 0.48 ± 0.02 for arousal, dominance, and valence, respectively) were significantly greater (Bonferroni-corrected p-value < 10-3) than the SVMR with the linear-kernel (0.18 ± 0.13, 0.12 ± 0.08, and 0.07 ± 0.11) and the non-linear-kernel (0.11 ± 0.14, 0.02 ± 0.12, and 0.05 ± 0.13). Overall, the DNNR features showed the strong negative intensities in the auditory areas, whereas these were mixture of positive/negative intensities such as in the anterior cingulate cortex, insula, and orbitofrontal cortex. 【Conclusions】 The performance of the DNNR was superior to the SVMRs in both (a) the automatic extraction of distinct features associated with human emotional processing and (b) the prediction of emotion scores based on the combination of these distinct features at the output layer.

この発表はCNNとLSTMによる皮質反応予測を行う発表でした．機能自体をDeepLearningでモデル化するというアイデアは非常に勉強になりました．

発表タイトル       ：　Sparse coupled hidden Markov models to probe temporally overlapping functional network interactions 著者                  ： Thomas Bolton セッション名           ： ORAL SESSION: Connectivity Methods and Analysis Abstruct ： 【Introduction】 The brain is active even in the resting-state (RS), as known from functional magnetic resonance imaging (fMRI) studies. Recently, the non-stationary nature with which its functional networks evolve and interact over time has been unraveled [1,3]. Using state-of-the-art deconvolution and clustering [4,5], this complex spontaneous brain activity can be disentangled into a set of interacting innovation-driven co-activation patterns (iCAPs), which map to known resting-state networks and show overlapping temporal activity profiles. To date, temporal modelling of network features is typically achieved through hidden Markov models (HMMs) [2,6], and interactions between networks have only been probed at the lower temporal resolution of second-order connectivity estimates, from non-deconvolved fMRI data [7]. To extend this exploration to the iCAPs, we introduce a sparse coupled HMM (SCHMM) framework enabling a sparse set of cross-network interactions. We first validate its implementation on artificially generated data, and then demonstrate the widespread presence of such modulatory influences in a RS fMRI dataset of healthy subjects. 【Methods】 In our framework, each iCAP lies in one of three possible hidden states at each time point: deactive, baseline, or active, with its temporal dynamics parameterized by the transition probabilities across those states (Fig. 1A). Further, (de)active iCAPs can temporarily modulate the transition probabilities of the other networks (Fig. 1B). Formally, the transition probability of iCAP k from state i to state j at time t depends on the activity state of the other networks at time t, as described by a logistic regression (Fig. 1C). To impose a physiologically plausible sparsity in modulatory influences, L1-regularization is casted on the logistic regression coefficients. To retrieve only significant modulations, following the determination of optimal regularization parameters, we perform comparison to null data with dismantled causality (independently circularly shifted network time courses; Fig. 1D). For validation purposes, we compared the accuracy of our SCHMM framework to a parallel HMM (PHMM) approach, in which cross-network interactions are not modeled, on an artificially generated system of three networks (20 sets of time courses, 1’000 samples) where network 2 exerts a modulatory influence on network 1.For real data analyses, we considered RS recordings from 20 healthy volunteers (38.4±6 years old) acquired with a Siemens 3T Trio TIM scanner, using a 32-channel head coil and gradient-echo echo-planar imaging (TR/TE/FA=1.1s/27ms/90º, matrix=64×64, voxel size=3.75×3.75×5.63mm3, 21 slices). We analysed time courses of 264 volumes for 9 iCAPs (Fig. 2C) showing extensive state transitions, and previously extracted in [5]. 【Results】 On artificial data, transition probability estimates for network 1 with and without modulation by network 2 were more accurate with the SCHMM than the PHMM approach (Fig. 2A). The same was observed for transition probabilities of the two other unmodulated networks (Fig. 2B). On real data, 28.9% of all possible modulatory influences were significant, and involved all examined iCAPs. Contrasting a condition with no cross-network coupling (intrinsic transition probability) to one where all possible modulations are incorporated (Fig. 2D), the primary visual and auditory networks showed down-regulated activity (larger probabilities for active-to-baseline switches and for remaining in the deactive state). Conversely, the precuneus/thalamus iCAP was up-regulated in activity (larger probabilities for baseline-to-active switches and for remaining in the active state). 【Conclusions】 Through explicit modeling of cross-network couplings by a SCHMM framework, intrinsic state transition dynamics could be successfully disentangled from external modulatory influences across some of the key resting-state brain networks. In particular, those modulations were shown to lower the activity level of sensory networks.

この発表はスパース隠れマルコフモデルを用いてネットワーク特徴の時間的モデリングをする発表でした．検証のために使用された人工的なデータセットを使っていて，私もモデルの検証時にこのようなデータセットを使うべきだと感じました．
 

発表タイトル       ：　Sharing deep generative representation for perceived image reconstruction from human brain activity 著者                  ： Changde Du セッション名           ： ORAL SESSION: Perception & Attention Abstruct ： 【Introduction】 Decoding human brain activities via functional magnetic resonance imaging (fMRI) has gained increasing attention in recent years. While encouraging results have been reported in brain states classification tasks, reconstructing the details of human visual experience still remains difficult. Two main challenges that hinder the development of effective models are the perplexing fMRI measurement noise and the high dimensionality of limited data instances. Existing methods generally suffer from one or both of these issues and yield dissatisfactory results. In this paper, we tackle this problem by casting the reconstruction of visual stimulus as the Bayesian inference of missing view in a multiview latent variable model (Fig. 1). Sharing a common latent representation, our joint generative model of external stimulus and brain response is not only “deep” in extracting nonlinear features from visual images, but also powerful in capturing correlations among voxel activities of fMRI recordings. The nonlinearity and deep structure endow our model with strong representation ability, while the correlations of voxel activities are critical for suppressing noise and improving prediction. 【Methods】 We present a deep generative multiview model (DGMM), where we cast the reconstruction of perceived image as the Bayesian inference of the missing view. Sharing a common latent representation, DGMM allows us to generate visual images and fMRI activity patterns simultaneously. For visual images, we explore nonlinear observation models parameterized by deep neural networks (DNNs), which can be multi-layered perceptrons or convolutional neural networks. This nonlinearity and deep structure endow our model with strong representation ability. For fMRI activity patterns, we adopt a full covariance matrix for the Gaussian distribution of voxel activities. While the full covariance matrix has the advantage of capturing the correlations among voxels, it results in severe computational issues. To reduce the complexity, we impose a low-rank assumption on the covariance matrix. This is beneficial to suppressing noise and improving prediction performance. Furthermore, we devise an efficient mean-field variational inference method to infer the latent variables and the model parameters. To further improve the reconstruction accuracy, the latent representations of testing instances are enforced to be close to that of their neighbours from the training set via posterior regularization. Compared with the non-probabilistic deep multiview representation learning models [5], our Bayesian model has the inherent advantage of avoiding overfitting to small training set by model averaging. 【Results】 We conducted experiments on three public fMRI datasets obtained from [1],[2] and [3]. We compared our DGMM method with the following algorithms: a specially designed method to reconstruct visual images by combining local image bases of multiple scales (Miyawaki) [1]; a Bayesian extension of CCA model that relates the fMRI activity space to the visual image space via a set of latent variables (BCCA) [4]; a latest deep multi-view representation learning model (DCCAE) [5]; a latest neural decoding method based on deconvolutional neural network (De-CNN) [6]. Extensive experimental comparisons demonstrate that our approach can reconstruct visual images from fMRI measurements more accurately than state-of-the-arts (Fig. 2). 【Conclusions】 We have proposed a deep generative multiview framework to tackle the perceived image reconstruction problem. In our framework, multiple correspondences between visual image pixels and fMRI voxels can be found via a set of latent variables. We also derived a predictive distribution that succeeded in reconstructing visual images from brain activity patterns. Although we focused on visual image reconstruction problem, our framework can also deal with brain encoding tasks. Extensive experimental studies have confirmed the superiority of the proposed framework.

この発表は視覚画像に対して，Deep Learningによって画像を再構成するときの特徴表現と，脳活動データを再構成するときの特徴表現を共有することで，画像とfMRIボクセルの対応関係を検討するという発表でした．パラメータ数が多いfMRIデータに対して，このような特徴表現のパラメータを使い回す手法を私も試すべきであると感じました．
 

発表タイトル       ：　Deep learning reveals brain features associated with preterm birth and perinatal risk factors 著者                  ： Manuel Hinojosa Rodriguez セッション名           ： ORAL SESSION: Lifespan Development Abstruct ： 【Introduction】 In the past decade, magnetic resonance Imaging (MRI) has been used at an increasing rate to study gray matter (GM) abnormalities in the brains of infants and children with a history of preterm birth and with perinatal risk factors for brain injury. Such abnormalities can occur in any region of the encephalic GM, though subcortical regions and the cerebellum tend to be most affected.2 GM abnormalities (GMAs) are usually stratified according to spatial distribution and to their degree of severity.2,3 However, clinical detection of mild-to-moderate abnormalities is visually challenging and may be insufficiently reported in clinical practice.4 The aim of our study is to apply machine learning (ML) an deep learning (DL) using neural networks (NNs) to distinguish between (A) pre-term and full-term children, and (B) children with various risk factors which are associated with MRI-detectable clinical conditions. 【Methods】 A total of 607 infants and children (326 males; age: 3.25 ± 2.22 years, μ ± σ) were recruited and participated in the study, which was compliant with the requirements of the Declaration of Helsinki. MP-RAGE T1-weighted MRI volumes were acquired sagittally at 3 T (voxel size = 1 mm3). Images were anonymized according to HIPAA requirements. In every subject, FreeSurfer 5.5 was employed to compute the volume, surface area, cortical area and the mean curvature of each gyrus and sulcus of the brain. This resulted in 592 structural brain features which were then used as input to a NN implementation of supervised ML. To avoid feature co-linearity, dimensionality reduction was implemented and 10 significantly uncorrelated features (Pearson’s r < 0.1, p < 0.05) were selected. Children were divided into three groups, according to whether they had (1) early pre-term, (2) late pre-term or (3) post-term births [i.e. born, respectively, (1) before 30 weeks, (2) 30 and 37 weeks, or (3) after 37 weeks of gestation]. According to this classification, there were 103 pre-term and 138 post-term births in the sample. Sex, gestational age and chronological age at MRI scan time were additionally included as NN feature variables to classify each subject as belonging to one of 7 groups, namely patients with (1) hypoxic-ischemic (HI) risk factors; (2) moderate-to-severe neonatal hyperbilirubinemia (NH); (3) both (1) and (2); (4) stroke; (5) malformations or genetic pathologies; (6) other abnormal conditions, and (7) healthy subjects. The limited-memory Broyden-Fletcher-Goldfarb-Shanno (BFGS) algorithm5 was used in conjunction with a NN architecture which involved hidden layers to classify nonlinearly separable data (L2 regularization penalty parameter = 0.001; random state=1; activation function: logistic). The number of hidden layers and of NN neurons were optimized to improve classification and avoid both over- and under-fitting (classification based on gestational age: 2 layer, 9 neurons; classification based on medication condition: 200 layers, 82 neurons). To ensure classification accuracy, 10 cross validation (CV) folds were implemented for each iteration. 【Results】 The brain features identified as having classification abilities were sulci associated with temporal and occipital regions bilaterally, namely the orbital, inferior temporal, middle occipital/lunate, posterior transverse collateral, the lateral occipito-temporal, collateral and lingual sulci. The NN was able to distinguish (A) children with pre- or post-term births from those with full-term births with a classification accuracy of 60.6% ± 0.7% (μ ± σ) and (B) among all seven sub-groups in the second classification with an accuracy of 75.6% ± 2.9%. 【Conclusions】 Our results suggest that certain brain features of very young subjects can be associated with aspects of developmental pathology. Thus, NNs and DL are promising methods for identifying relationships between brain structure and medical conditions which affect early neurodevelopment.

この発表は幼児の灰白質の異常を検知するために，Deep Learningを用いるという研究でした．この発表を聞いて私は，目視で分かる画像に対して，この異常検知を自動化する理由はあるのかという疑問を持ちました．
 
学会参加報告書

報告者氏名	中村圭佑
発表論文タイトル	SLICとNormalized Cutを用いた脳領域の分割手法
発表論文英タイトル	Brain region segmentation method using SLIC and Normalized Cut
著者	中村圭佑，日和悟，廣安知之
主催	Organization for human brain mapping
講演会名	OHBM 2017
会場	Vancouver convention centre
開催日程	2017/6/25-2017/6/29

 
 

1. 講演会の詳細

2017/6/25-29に，カナダのバンクーバーにて開催されましたOHBM 2017に参加いたしました．OHBM2017 は，Organization for human brain mappingによって主催された国際会議で，ヒトの脳組織に関する研究に携わる様々な背景の研究者を集め，これらの科学者のコミュニケーション，および教育を促進することを目的に開催されています．本研究室からは他に廣安先生，日和先生，池田さん，石田翔也くん，片山さん，石原さん，萩原さん，玉城さん，吉武さん，相本くん，藤井聖香さん，三好くん，水野さんが参加しました．
 

2. 研究発表
   • 発表概要

私は26日のポスターセッションに参加いたしました．発表の形式はポスター発表で，2時間自由に参加者の方と議論を行いました．
今回の発表は，「Brain region segmentation method using SLIC and Normalized Cut」と題して発表いたしました．以下に抄録を記載します．

Introduction：In the brain function analysis, identification of ROI (Region Of Interest) is critical in considering the activation site and functional connectivity of the brain. The shape and location of the ROI in the brain is defined by the brain atlas. However, most of the brain atlas are those labeled based on the brain anatomical division of the brain part, or regions labeled with functional connections. Therefore, in the former brain atlas, the number and size of ROIs are arbitrarily changed, and it is hard to examine the brain function. Also, since the shape of the brain varies among individuals, it is difficult to consider the difference in brain shape in the method of determining the ROI using the conventional atlas. In this study, we developed an automatic brain atlas creation system that can change the number of region divisions. Methods：In the proposed method, structural images and functional images by MRI (Magnetic Resonance Imaging) are segmented by SLIC (Simple Linear Iterative Clustering) and Normalized Cut [. In this method, the number of divisions of the brain region is changed by changing parameters at the time of image division. To confirm the effectiveness of the method, the proposed method was applied to the T1 image on which the standard cerebralization process was performed. The T1 images of two men were used, and the division of the obtained brain regions was compared. For quantification of regional division, the similarity was calculated by Jaccard Index. Results：By changing the parameters of SLIC and Normalized Cut, about 150-300 regions were divided. The results were good division results along the anatomical division.The average value of similarity of the region shape of the two subjects was about 23% when divided into 210 regions. At this time, the similarity of the region shape was 56% at the maximum and 3% at the minimum. Therefore, the results of region segmentation of this method showed that differences occur depending on the position of the brain region. Conclusions：In this study, a brain atlas creation system based on brain region division by arbitrary division number using SLIC and Normalized Cut was proposed. The proposed method was applied to the T1 image of two subjects, and the division result by the Jaccard Index was quantified. As a result of region segmentation by the proposed method, the T1 image was divided into 150 to 350 regions. According to the Jaccard Index of the region shape after the division of the two subjects, regions with high similarity and regions with low similarity existed.

 

• 質疑応答

今回の講演発表では，以下のような質疑を受けました．
 
・質問内容1
複数名の方からSLICの分割で用いている初期クラスタのKの値はいくらかという質問をいただきました．こちらの質問に対する回答ですが，今回のケースでは1024領域に分割しているが，もしより細かい分割が必要であるならば，Kの値を大きくする必要がある，と回答いたしました．
 
・質問内容2
Fiber Trackingの際に用いているパラメータについての質問をいただきました．Fiber Trackingを行う際のパラメータを控えていなかったため，その旨を返答致しました．今後のこのようなことが無いよう，使用したツール・機器の設定/パラメータを控えておくように心がけます．
 
・質問内容3
評価に用いているDice係数の値が低いが，個人のデータごとに分割を行う際にどのような要因によりクラスタリング結果に違いが生じているのかという質問をいただきました．この質問に対し，Fiber Trackingの結果が個人により異なるため，生成されるアトラスも異なる分割となっている．と答えました．
 

• 感想

今回は初めての学会発表であり，また初の英語発表ということで，自身の発表をどこまで伝えられるか，自身の研究のどこをアピールできるのか，ということを考えて準備してきました．今回のポスター発表では，ポスターのファーストインプレッションという点ではセッションの最初であまり人の目を引くことが出来ず，ポスターの見た目のインパクトと題目の重要さを身をもって痛感し，研究のアピールという点では不十分であったと思います．この点は，今後の発表で活かしたいと思います．一方，セッション終盤では自分と同じParcellationの分野に携わっている方々と意見を交わすことができ，自分の研究を十分に伝え，かつ有意義な議論ができたと感じました．また，その他のParcellationの研究やAtlas basedアプローチを行っている研究について聴講し，脳機能研究における自分の研究の立ち位置を把握することが出来，今後の自分の研究に励みになったと考えています．
 

3. 聴講

今回の講演会では，下記の5件の発表を聴講しました．
 

発表タイトル      ：　Performance of Various Brain Atlases for Individual Idetification using resting fMRI 著者              ： Andrew Michael, Chao Zhang セッション名       ： Oral Session(Informatics) Abstruct        ： Introduction: In a recent paper (Finn et al. 2015) it was demonstrated that functional connectivity (FC) features based on resting fMRI (rfMRI) data can be used as a “fingerprint” to accurately identify individual subjects. Subsequent studies have since demonstrated the utility of rfMRI in predicting individual differences in a wide array of cognitive measures such as intelligence (Hearne, Mattingley, and Cocchi 2016), distractibility (Poole et al. 2016), and attention (Rosenberg et al. 2015). The finding that rfMRI can be utilized to predict individual differences has paved the way for the potential use of functional “fingerprints” in the treatment and diagnosis of psychiatric disorders. However, the above studies were mostly based on a small number of subject (N~100). In this study we address the following questions: (1) can rfMRI FC predict individuals in a large cohort? (2) what brain atlases are the best performers for individual identification? (3) can rfMRI data acquired on a different day be used for individual prediction? and (4) what is the separation between the best identification and the next best match? Methods: Our study contained rfMRI from 820 healthy young adults (366 males and 454 females, age: 22-37 years) from the Human Connectome Project (HCP) S900 release (Van Essen et al. 2012). Each subject was scanned on four different runs (2 each on 2 separate days). For each run and each subject, time series information was extracted from ROIs as defined by the following ten different brain atlases: DOS160 (Dosenbach et al. 2010); CC400/CC200 (Craddock et al. 2012); AICHA (Joliot et al. 2015); Stanford90 (Shirer et al. 2012); Harvard-Oxford; Automated Anatomical Labeling (AAL); AAL_new; AAL2; and Brodmann. Individual FC was calculated between ROI time series using Pearson correlation. To implement individual identification, we correlate the FC of each subject from Run1 to FC of all 820 subjects from Run2 and paired subjects based on maximum correlation. The prediction accuracy is defined as the proportion of subjects with correct identification. We then repeat this process to identify subjects using Run3 and Run4 data. Results: Prediction accuracies for the ten different atlases are presented in Figure 1. DOS160 produced the highest accuracy of 95%. The accuracies of four other functional atlases were above 80%. We note that the performance of the five structural atlases was in the range of 54–66%. Prediction accuracies for Run3 and Run4 data were 95% and 88% respectively (Figure 2). In Figure 2 we further investigate the FC correlation of the correctly identified subjects (in red) and the next best 20 FC correlations (in blue). We note that for a large proportion of the 820 subjects, the second best match is significantly lower than the correct match indicating the robustness of rfMRI FC for individual identification. Conclusions: We performed individual identification using rfMRI data for a large cohort of 820 subjects and show that the DOS160 atlas is the best performer. Of the atlases examined, the five functional parcellations demonstrate much higher identification accuracies (above 80%) than the five structural parcellations (<66%). We show that high prediction accuracies are possible between rfMRI data acquired on different days. We conclude that choice of parcellation scheme is an important consideration for studies performing individual identification. By improving characterization of FC differences at the individual level, it may be possible to gain novel insights into the association between individual FC differences and distinct cognitive or behavioral features.

本発表は，脳の機能的コネクティビティを用い，”figer print”として個人を同定することを目的とするものでした．発表においては，用いるアトラスによって同定の精度が異なるという結果が示されており，機能的コネクティビティを考慮する際におけるアトラスの重要性に気づかされました．
 

発表タイトル      ： Brainnetome Atlas: A New Map of Human Brain 著者              ： Lingzhong Fan, Hai Li, Zhengyi Yang, Tianzi Jiang セッション名       ： Oral Session(Informatics) Abstruct        ： Introduction: The brain atlases based on different mapping techniques are the navigator of the human brain, and considered as the cornerstone of basic and clinic neuroscience(Toga et al., 2006; Evans et al., 2012; Amunts & Zilles, 2015). With a history of more than a century, the Brodmann’s map developed by a neuroanatomist, Korbinian Brodmann, divided the human cerebral cortex into 52 different areas based on its cellular architecture, is still used most often as one of the possible parcellations(Zilles & Amunts, 2010). However, the limitations of this map have become more and more obvious, increasing the importance of defining brain areas using new methodologies(Paxinos, 2016). In the year of 2010, the Brainnetome project was launched to investigate the hierarchy in human brain from genetics and neuronal circuits to behaviors. One of the key elements of this project is focused on setting up and optimizing the framework for connectivity-based parcellation, and aims to produce a new human brain atlas, i.e. Brainnetome atlas based on connectional architecture (Jiang, 2013; Fan et al., 2016).Currently, the human Brainnetome Atlas is freely available for download at http://atlas.brainnetome.org, so that whole brain parcellations, connections, and functional data will be readily available for researchers to use in their investigations into healthy and pathological states. Methods: Using noninvasive multimodal neuroimaging techniques, we designed a connectivity-based parcellation framework that identifies the subdivisions of the entire human brain, revealing the in vivo connectivity architecture. The resulting human Brainnetome Atlas, with 210 cortical and 36 subcortical subregions, provides a fine-grained, cross-validated atlas and contains information on both anatomical and functional connections. Additionally, we further mapped the delineated structures to mental processes by reference to the BrainMap database. As part of this work, we developed an integrated “Automatic Tractography-based Parcellation Pipeline (ATPP)” to realize the parcellation using automatic processing and massive parallel computing (Fig. 1) that we share with the atlas. Results: This new brain atlas has the following four features(Fig.2): (A) It establishes a fine-grained brain parcellation scheme for 210 cortical and 36 subcortical regions with a coherent pattern of anatomical connections; (B) It supplies a detailed map of anatomical and functional connections; (C) it decodes brain functions using a meta-analytical approach; and (D) It is an open resource for researchers to use for the analysis of whole brain parcellations, connections, and functions.The Brainnetome Atlas together with its related software is available for download to serve as a shared community resource. The pipeline software is open to the community to facilitate the parcellation of specific brain regions of interest. Conclusions: The human Brainnetome Atlas could constitute a major breakthrough in the study of human brain atlas and provides the basis for new lines of inquiry about the brain organization. It will enable the generation of future brain atlases that are more finely, defined and that will advance from single anatomical descriptions to an integrated atlas that includes structure, function, and connectivity, along with other potential sources of information. It will present neuroscientists with one of the key tools that will help us get some entirely new knowledge on how the brain works, as well as to understand the pathophysiological mechanism of psychiatric and neurological disorders.

 
本発表は，機能的コネクティビティ，および神経追跡を用いて脳アトラスを作成する手法についての発表でした．上記のデータを用いてデータドリブンなアプローチでアトラスを作成する手法であり，2016年に発表されたものでしたが，OHBM中にこのアトラスを用いた研究を複数みかけました．自身のアプローチに近い手法であるため，今後どのようにアトラスを作っていくかを考えさせられました．
 

発表タイトル      ：Adaptive Cortical Parcellaions for Source Reconstructed EEG/MEG Connectomes 著者              ：Seyedehrezvan Farahibozorg, Richard Henson, Olaf Hauk セッション名       ： Oral Session(Modeling & Analysis) Abstruct        ： Introduction: There is growing interest in the rich temporal and spectral properties of Electro- and Magnetoencephalography (E/MEG) signals in order to study the functional connectome of the brain [1, 2]. However, the spatial resolution of E/MEG data is limited, because several thousand sources of activation in the brain must be estimated from maximally a few hundred recording sites. This limited spatial resolution causes the so-called leakage problem: activity estimated in one region of interest (ROI) can be affected by leakage from locations outside this ROI [3, 4]. E/MEG studies typically adopt parcellations from structural or fMRI research for whole-brain connectivity analysis [5]. However, considering the spatial resolution of E/MEG, these parcellations are unlikely to be optimal [6]. Here, we utilise Cross-Talk Functions (CTFs) as a direct measure of spatial leakage [7] and utilise two CTF-informed image segmentation algorithms in order to parcellate the cortical surface into the maximum number of distinguishable ROIs. Methods: We computed resolution matrices (with rows as CTFs) for individual subjects, based on forward and inverse models computed using BEM head models and L2 MNE inverse operators of 17 healthy subjects. In the first parcellation approach, we started from standard anatomical parcellations and modified the ROIs using a CTF-informed split-and-merge (SaM) algorithm [8]. In the second approach, we started from all brain vertices with no prior parcellation. A CTF-informed region growing (RG) algorithm [8] was used to create ROIs around the vertices that showed highest sensitivity and specificity of CTFs on the cortex, which were then optimised using an SaM algorithm. The algorithms are designed such that they merge ROIs/vertices with highly overlapping CTFs, split ROIs that produce distinguishable patterns of CTFs, remove ROIs with low sensitivity, and for each ROI identify a group of representative vertices that show high sensitivity and specificity to that particular ROI. We used ROI Resolution Matrices (RRmat) to quantify leakage from each ROI to all other ROIs in the brain in order to evaluate the parcellations’ performance where an ideal RRmat is an identity matrix. Thereafter, we evaluated the possible consequences of using different parcellation methods for graph-theoretical connectivity analyses on simulated data with realistic levels of noise. Results: Based on the RRmats (Fig. 1), we found that parcellation sensitivity improved from 0.47 and 0.37 in two standard anatomical parcellations (Desikan-Killiany (DKA) and Destrieux Atlases (DA) respectively) to 0.65, 0.70 and 0.70 in modified DKA, DA and RG parcellations respectively. Moreover, ROI distinguishability improved from 0.50 and 0.38 to 0.61, 0.65 and 0.64 (Fig. 1). Interestingly, in spite of their different starting points, both SaM and RG algorithms yielded approximately 70 ROIs. Furthermore, our simulated realistic connectome with a single hub showed that modified parcellations were particularly successful in improving hub sensitivity and hub connectivity probability patterns (Fig. 2). Conclusions: Our proposed parcellation algorithms significantly improved the sensitivity and distinguishability of ROIs compared to the anatomical parcellations, while at the same time maximising the number of distinguishable ROIs in the brain. The algorithms are adaptive with respect to the measurement configuration and source localisation methods. Regardless of the starting point they yielded around 70 ROIs, suggesting that this reflects the resolution limit of this particular sensor configuration and source estimation method. Furthermore, our simulations showed that the choice of parcellation can have significant impact on the outcome of graph theoretical analysis of the source-reconstructed E/MEG. Therefore, we conclude that adaptive parcellations are essential for whole-brain EEG/MEG connectomics.

 
本発表は，EEGおよびMEGを用いて機能的コネクティビティを考慮する際に発生するleakage problemに対して，用いるアトラスの形状とサイズをCross-Talk FunctionによってE/MEGにとって最適なものにすることで，leakage problemの解決を試みる発表でした．NIRSについても同様のことが出来そうな内容であり，考慮する必要性を感じました．
 

発表タイトル      ：　Unravelling the intrinsic functional boundaries of the macaque monkey cortex 著者              ： Ting Xu, Alexander Opitz, Arnaud Falchier, Gary Linn, Deborah Ross, Julian Ramirez, Darrick Sturgeon, Eric Feczko, Elinor Sullivan, Jennifer Bagley, Stan Colcombe, Damien Fair, Charles Schroeder, Michael Milham セッション名       ： Oral Session(Modeling & Analysis) Abstruct        ： Introduction: A growing body of literature has demonstrated the ability to delineate cortical areas in the human brain based upon the detection of spatial transitions in intrinsic functional connectivity (iFC) profiles (Cohen et al., 2008; Wig et al., 2014). In particular, gradient-based parcellation approaches have gained popularity due to their ability to recapitulate previously established cytoarchitectonic brain areas. Here, we demonstrate the feasibility of extending the application of parcellation approaches to non-human primates (NHP), demonstrating the reliability of these parcellations and comparing the cortical areas revealed to those obtained in humans. Methods: We collected data from a male rhesus macaque monkey (age: 6 year) on a 3 Tesla Siemen Tim Trio scanner. Awake functional MRI scans were obtained during 6 sessions (4-7 scans for each session, 8 minutes per scan, 216 minutes in total, TR = 2 s, 1.46 x 1.46 x 2 mm); three of the sessions were carried out using a contrast agent (i.e., monocrystalline iron oxide particle (MION)) and 3 were without contrast. We obtained high-resolution T1-weighted anatomical images (0.5mm isotropic voxel) for surface registration. The native surface was reconstructed and registered to Yerkes19 macaque template (Donahue et al., 2016). We calculated iFC-similarity maps for each scan, followed by the spatial gradient and edge detection computation on native surface. The spatial correlations were calculated to investigate the reproducibility of boundaries across sessions and scans. We further explored the requirement of scan time for a relatively robust iFC and boundary map. Results: As expected, whole-brain gradient maps exhibited a higher degree of similarity among individuals within the same developmental period; differences were particularly notable at the extremes (i.e., childhood, older age) (see Figure 1A). To facilitate visualization, we defined 6 age groups and depicted mean gradient maps in Figure 1B. Next, at each voxel, we used univariate analyses to detect age-related linear and quadratic trends in global mean for the gradient map associated with that specific vertex. These analyses revealed linear age effects in posterior cortex, particularly in primary visual, sensorimotor, and default mode networks (Figure 1C). The quadratic effects were mainly located in the regions of network borders e.g. default mode, ventral attention (Figure 1C). Finally, at each vertex, we used MDMR to detect age-related variation (linear, quadratic) in the gradient maps defined across individuals. The linear and quadratic age-related effects were predominantly located in the regions of network borders, e.g. default mode, ventral attention, dorsal attention and frontoparietal network (Figure 2). Conclusions: By examining the transition pattern of iFC similarity in macaque, we have demonstrated the ability to detect functional boundaries and cortical areas in the macaque monkey cortex using awake R-fMRI in macaque, suggesting a reliable scheme for delineating cortical organization in macaque and potential utility for validating invasive individual parcellation.

 
本発表は，マカクザルの脳の機能的境界をアトラスを用いて定義し，そのアトラスを用いてマカクザルの脳機能を解析する，といった内容でした．アトラス作成に用いられたgradient-based parcellationと評価の方法については，今後の研究で機能データを用いる際の参考になると思いました．
 

発表タイトル      ： Predicting Personality from Network-based Resting-State Functional Connectivity 著者              ： Alessandra Nostro, Veronika Müller, Deepthi Varikuti, Rachel Pläschke, Robert Langner, Simon Eickhoff セッション名       ： Oral Session(Social Neuroscience) Abstruct        ： Introduction: Personality as a key feature of inter-individual differences affects all aspects of life, including affective, social, executive and memory functioning [3,4,6]. Task-based fMRI studies investigated personality and brain activity in association to each of these domains; however, since personality traits are enduring across situations [2], it is possible that they relate to many brain systems, not detected by task-based fMRI. The investigation of functional connectivity in resting state conditions might therefore help in capturing the intrinsic and complex neural architecture underlying personality [1]. A recent study [7] showed a sexual dimorphism in brain structure-personality relationships, with associations revealed only in males. In females, brain connectivity rather than structure, might thus play a stronger role in light of personality. Therefore, we aimed to predict scores of the five-factor personality model (openness, conscientiousness, extraversion, agreeableness, neuroticism) [2] from resting-state functional connectivity (RS-FC) in meta-analytically defined brain networks, and tested how these predictions are modulated by gender. Methods: We assessed 9 meta-analytic networks representing regions consistently activated by different social (empathy, face perception), affective (reward, pain, emotion perception), executive (working memory, vigilant attention) and mnemonic (autobiographic and semantic memory) functions. FIX-denoised RS fMRI data of 136 males and 137 matched females was downloaded from the HCP WU-Minn Consortium [10] and further preprocessed with SPM8 using standard procedures. Within each network, FC between all nodes was computed using their respective extracted time series. A relevance vector machine-learning algorithm [9] was used to predict NEO-FFI scores [2] based on FC between all nodes of each network, separately for males and females. Prediction performance was assessed by Pearson correlations between real and predicted scores (p<0.05, corrected for multiple comparisons) and compared between groups. Results: Personality traits were successfully predicted by FC within different networks in men and women (see Fig. 1 for a summary). Specifically, in men, conscientiousness was predicted by FC within networks of the affective system (e.g. r=.40 for the reward network; Fig. 2A), extraversion by networks related to social, memory and affective processing, and agreeableness by networks of affective and social domains. In women, openness was predicted by FC within affective and memory-related networks (e.g. r=.45 for the autobiographic memory network; Fig. 2B), conscientiousness by networks linked to executive functioning, and neuroticism by memory-related network. Significant gender differences in prediction performance were found for openness, conscientiousness and agreeableness (Fig. 1). Conclusions: Using machine-learning techniques the current study revealed substantial associations of personality with various brain networks related to affective, social, executive, and long-term memory functions, based on FC within these networks. These results indicate that RS connectivity patterns within meta-analytically defined functional brain systems provide information on the individual expression of specific personality traits. Indeed, they were not only predicted by networks already associated to them in the literature, but also not expected brain systems were found informative, with the exception of neuroticism which was not predicted by any expected affective networks. Additionally, FC patterns of different functional networks were shown to predict different personality traits in males and females, indicating gender-specific neural mechanisms associated with specific personality characteristics. This extends previous findings on relations between network-specific differences in gray-matter volume and personality [7] by demonstrating that RS-FC–personality relations should not be considered independent of gender.

 
本発表は，resting stateにおける脳の機能的コネクティビティを解析することで，個人の性格を推定する，といった内容でした．性格の各要素に関連する脳の機能的ネットワークをmeta analysisを用いて抽出しており，今後の研究においてこのような手法を活用できるのではないかと思いました．
 
参考文献

• OHBM 2017,

https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageID=3734