【速報】OHBM2019

OHBM2019にて発表しました。

学会参加報告書

報告者氏名	杉野梨緒
発表論文タイトル	Behavioral and functional connectivity analysis of Kanizsa illusory contour perception
著者	Rio Sugino, Satoru Hiwa, Keisuke Hachisuka, Fumihiko Murase, Tomoyuki Hiroyasu
主催	Organization for Human Brain Mapping
講演会名	25th Annual Meeting of the Organization for Human Brain Mapping
会場	Auditorium Parco Della Musica
開催日程	2019/06/9-2019/06/14

 
 

1. 講演会の詳細

2019/06/9から2019/06/14にかけて，Auditorium Parco Della Musicaにて開催されました25th Annual Meeting of the Organization of Human Brain Mappingに参加いたしました．この25th Annual Meeting of the Organization of Human Brain Mappingは，Organization for Human Brain Mappingによって主催された国際会議で，ヒトの脳組織および脳機能のマッピングに関する研究に携わる様々な背景を持つ研究者を集め，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1）．
私は10～14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，古家，大塚，奥村，山本，吉田，風呂谷，丹が参加しました．
 

2. 研究発表
   • 発表概要

私は11日の午後のセッション「Poster session」に参加いたしました．発表の形式はポスター発表で，発表時間は1時間となっておりました．
今回の発表は，Behavioral and functional connectivity analysis of Kanizsa illusory contour perceptionでした．以下に抄録を記載致します．

Background: Illusory contour (IC) is one of the common illusions encountered in our daily lives. The neural basis involved in IC perception has been studied so far (Ritzl Afra et al 2003), but the relationship between behavioral response and brain activity has remained unexplored. In this study, we considered IC perception speed as a behavioral measure, and assumed a functional connectivity (FC) to be correlated with it. Kanizsa figures were used to induce IC perception, and a dot localization task was used to measure reaction time (RT) to the IC. Brain activity during IC perception was measured and functional connections correlated with RT to IC were extracted by FC analysis. Methods: Eighteen healthy adults (23.1±1.2 y/o, 4F/14M) were instructed to perform a dot localization task in an fMRI scanner (Chen Siyi et al 2018) . In the task, they judged whether the dots were inside or outside the contour, and pressed a button to answer. Fig.1 shows the experimental design. The dots were presented in a pseudo-random position with the same inside and outside probability. RT from the instance of target presentation to that of response was used as a behavioral metric reflecting the contour perception difference between the participants. The difference in average RT between IC and real contour (RC) tasks was used as an indicator of IC perception speed, and on the basis of the median value the participants were divided into FAST and SLOW groups. These two groups were compared to investigate how the difference in RT was reflected on FC network. The whole brain was parcellated into 116 regions based on the AAL atlas. A correlation coefficient matrix was calculated from region-of-interest-wise BOLD signal of each task using CONN toolbox. FC matrices for each group were compared between two tasks; and FC which was higher in the IC task than the RC task, was extracted and compared among the two groups. Furthermore, a correlation analysis was performed between FC of these connections and the difference in RTs between the two tasks. Results: Mean values of RT differences between IC and RC tasks of FAST and SLOW groups were 131.2±28.0 ms and 218.5±69.0 ms, respectively, and there was a significant difference between the two groups (p<0.05). Fig.2a shows functional-connection differences between IC and RC tasks, which significantly differed between the two groups (p<0.05, FDR). The red line indicates the connection whose difference between IC and RC tasks is higher in the FAST group than that in the SLOW group, while the blue line shows those with a higher difference in the SLOW group. In addition, Fig.2b shows a significant correlation between the functional connection differences and average RT differences (p<0.05). The functional connections between the right supplementary motor area (SMA.R) and the orbital parts of left/right superior frontal gyri (ORBsup.L/R) were the highest among the extracted connections. SMA is the region belonging to the salient network and is related to awareness (Power Jonathan D. et al. 2011), while ORBsup is related to top-down attention (Aboitiz Francisco et al 2014). Since FC is higher in IC task than RC task, it is conceivable that these connections are related to IC perception. Furthermore, it was suggested that higher connections among these regions in the FAST group than in the SLOW group could be associated with an increase in IC perception speed. Conclusions: In this study, FC correlated with IC perception speed was investigated using Kanizsa figures. Participants were divided into FAST and SLOW groups according to the IC perception speed, and FC was found to be higher in IC task when compared between the groups. Thus, it was shown that six functional connections in the FAST group had higher connectivity in the IC task than in the RC task. Among these connections, FC between the regions related to awareness and attention was high. These results suggest that a higher temporal synchronization is involved between these regions in IC perception.

 

• 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
正答率はどのタスクに対して算出しているのかについて質問を受けました．この質問に対して，ICタスクでの正答率を算出していると答えました．
・質問内容２
なぜ5本の結合のみが抽出されたのかについて質問を受けました．この質問に対して有意水準0.1%で無相関検定を行ったため，p値が0.001より小さい結合を抽出したためと答えました．
・質問内容３
FC解析は全脳で行われているのか，結合はいくつあるのかについて質問を受けました．この質問に対して，全脳は116領域に分割され，その全ての領域間でFCは算出されたと回答しました．
・質問内容４
RCタスクを基準として用いることについて質問を受けました．この質問に対してRCはICに対するコントロールタスクであるため，基準として用いたと回答しました．
・質問内容5
sharpnessが高い群と低い群の2つ群があるのかについて質問を受けました．この質問に対して被験者22名の中に高い人もいれば低い人もいると回答しました．
・質問内容6
考察の解釈について質問を受けました．この質問に対してSharpnessが低い人は正答率が低く，タスクを行うためによりエネルギーを使っているためFCが高くなると考えていると答えました．
・質問内容6
実験設計は事象関連デザインかについて質問を受けました．この質問に対してブロックデザインを用いていると回答しました．
・質問内容7
扁桃体が結果に含まれていることについて質問を受けました．この質問に対して視覚システムは扁桃体との構造的結合を持っており，扁桃体は情動に関わる領域であるため，情動の変化が感覚の知覚に関わるのではないかと解釈していると回答しました．
 
2.3.        感想
2回目の国際学会の参加だったため前回よりも落ち着いて発表することができた．今回は方法が比較的シンプルであったため，前回よりも正確に伝えることができたと感じている．良かった点は，学会を通して顔見知りになった方にポスターを聞きにきて欲しいと伝えたことにより，親身に発表を聞いてくれたことで，説明と質問を繰り返して研究の隅々まで伝えることができたことが挙げられる．反省点は名刺を要求されたが持っていなかったことが挙げられる．去年同様に英語を聞き取るのは苦手ではあったが，諦めずに聞き直し，自分で言い直し合っているかを確認しながら相手の発言を聞くことが出来たことができた．
 
 

3. 聴講

今回の講演会では，下記の４件の発表を聴講しました．
 

発表タイトル       ：Task-dependent functional organizations of the visual ventral stream 著者                  ： Han-Gue Jo, Thilo Kellermann, Junji Ito, Sonja Grün, Ute Habel セッション名       ： Poster session Abstruct            ： Background: The visual ventral stream is a series of hierarchical processing stages from the primary visual cortex V1 to inferior temporal cortex IT, in which neural interactions along this hierarchy enable us to recognize visual objects. However, its complex and diverse connectivity make it difficult to illustrate the functional organization, particularly when top-down cognition is involved. Depending on task-goal, the ventral stream may require different functional structure of the hierarchy to incorporate visual features of interest into object recognition [1,2]. Here we identified context-dependent functional structures of the ventral stream. Methods: Twenty-eight participants performed three types of visual cognition task during fMRI measurement. The three task conditions that required distinct cognitive processes for object recognition were used in order to drive the visual ventral stream: searching for a target object, memorizing objects in natural scenes, or free viewing of the same natural scenes. We identified a task-dependent connectivity network of the ventral stream, utilizing a hierarchical seed-based connectivity approach that explicitly compared task-specific BOLD time-series. Seed-based analysis was performed within the ventral stream, and the first cortical processing stage V1 was subjected as a seed region. Voxel clusters that revealed significant task effect were identified as regions of interest (ROIs) and these ROIs were further subjected as seeds for subsequent seed-based analyses. On the basis of the identified ROIs, we demonstrated task-dependent connectivity to which extent the connectivity increases or decreases during each of the visual search, memory, and free viewing conditions. Results: The hierarchical seed-based connectivity approach identified five ROIs in the visual ventral stream (Figure 1), representing a task-dependent functional network. The connections across the identified ROIs were organized into correlated and anti-correlated structures according to the context of visual cognition. Searching for a target object separated the visual area V1 and V4 from the high-order visual area PIT (the posterior part of the IT), while memorizing objects strengthened the coupling of V4 with PIT. Furthermore, task-dependent activation was found in V1 and V4, while the PIT showed deactivation. Conclusions: The present study demonstrated context-dependent functional structures of the visual ventral stream. In particular, while the ventral stream was organized into correlated and anti-correlated structures during searching for a target object, memorizing objects manifested a correlated structure. Our results further suggest a putative boundary between V4 and PIT, which divides the visual hierarchy into two subdivisions that interact competitively or cooperatively depending on task demand. These results highlight the context-dependent nature of the ventral stream and shed light on how the visual hierarchy is selectively mediated to bias object recognition toward features of interest.

この発表は腹側皮質視覚路のコンテキスト依存機能構造を識別に関する研究でした． 3 種類の視覚認識課題（目標対象物の探索，自然シーンにおける対象物の記憶，または同じ自然シーンの自由視聴）を用いることにより，腹側皮質視覚路の機能である物体特定や物体認識の働きを見ていました．この3種類のタスクが非常に興味深かったです．またこの視覚路は神経相互作用によって視覚的対象を認識することが可能であると言われており，先日調査したばかりだったので，また新たな知見を得ることができ非常に勉強になりました．解析においては，腹側皮質視覚路の処理段階の階層において1番初めの領域であるV1をSeed領域としてSeed-based analysisを行い，有意なタスク効果を持つことがわかった領域をROIとして同定し，さらにそのROIをSeed領域としてSeed-based analysisを行うという方法でした．この解析方法は真似てみることが出来そうだったので，今後の研究にも活かしたいと思っています．

発表タイトル       ：10,000 Social Brains: Charting sexual dimorphism in the UK Biobank 著者                  ： Hannah Kiesow セッション名       ： Oral session: Population Neuroscience Abstruct            ： Reliance on one’s social network provides many advantages. The social brain hypothesis (Byrne & Whiten, 1988; Humphrey, 1976) argues that neocortex volume in primates co-evolved with the cognitive costs required to maintain complex social environments. Various complex social indices have structural implications in the brain. For example, larger social network size was associated with increases in gray matter volume in regions engaged in social processing in the human (Lewis, Rezaie, Brown, Roberts, & Dunbar, 2011) and nonhuman (Sallet et al., 2011) primate brain. Individual variation in social brain volumes may be expressed according to sex. Sex has been argued to be the phenotypical distinction that explains most behavioral variability in most species. The different behavioral profiles of males and females presumably rely on distinct topographical brain circuits that are anatomically or functionally dimorphic. Thus, the aim of our study was to investigate the neural manifestations of sexual dimorphism in different indices of social behavior.

この発表は10000人の分の脳構造データを用いた性差研究でした．性別はほとんどの種において，もっとも行動の多様性を説明する表現型の違いであると主張されており，男性と女性の異なる行動のプロファイルは解剖学的または機能的に異なる脳回路に依存していると考えられている．そのため社会的行動の異なる指標（社会的内輪指標，社会的アウトレット指標，社会的経済指標，社会文化的移行指標）を用いて性的二形性の神経基盤について研究がされていた．解析として社会的指標の脳容積の分布を明示的にモデル化する確率的階層モデリングを用いていた．このモデルは予想される社会人口統計学的変動を表すために年齢と性別に条件づけられており，長期可塑性効果の結果を反映している．その結果，社会的な脳の特定の領域の体積と関連していること，機能的結合ネットワークを通して違いが検出されたこと，類似した生活習慣因子の社会的指標の中で識別できる複雑な社会変数における性関連の影響が示唆されたと発表していた．私たちの研究室でもテーマとして用いられていることから性差に関する研究は身近であるものの，これほどのビッグデータに対して研究が行われていることに驚きました．しかし最近調査を行ったサンプルサイズ設計を考慮すると，サンプルサイズが大きいことのデメリットが存在するのではと感じました．普段は脳機能研究の発表を中心に聴講しているので，脳構造研究の話は非常に新鮮で興味深いと感じました．
 
 
 

発表タイトル       ：Changes in pRFs during perceptual filling-in of an artificial scotoma in humans 著者                  ： Joana Carvalho, Remco Renken, Frans cornelissen セッション名       ： Poster session Abstruct            ： Background: When the information extracted from a visual scene is incomplete, the visual system attempts to predict what is missing by extrapolating from nearby information. This is filling-in. Despite its clinical and scientific relevance, the neuronal mechanisms underlying filling-in are still ill-understood. Here, we used fMRI in combination with population receptive field (pRF) mapping to determine how and where in the visual hierarchy filling-in takes place. In our experiment, we measured pRF properties during and in the absence of perceptual filling-in, while observers viewed band-pass filtered textures on which artificial scotomas could be superimposed. Methods: Seven observers (3 females; age-range: 26–32) with normal or corrected to normal vision were scanned using a Siemens Prisma 3T scanner. Retinotopic mapping was performed using:1) spatial frequency retinotopy (SFR) – the contrast between the carrier and the bar was only perceived on the basis of spatial frequency; and 2) SFR with four artificial scotomas superimposed (SFR_scot). The scotomas were centred at each quarter field at 5 deg eccentricity, the diameter of the scotomas was 3 deg, as depicted in figure 1A. High contrast localiser scans were used to obtain the locations of the artificial Scotoma Projection Zones (aSPZ). During scanning, participants were asked to perform a fixation task: they had to press a button each time the fixation point changed from red to green. For both SFR and SFR_scot, a single run consisted of 136 functional images (204 s). The pRF estimation was performed using the mrVista (VISTASOFT) Matlab toolbox and using a custom implementation of Bayesian pRF. Data was thresholded by retaining the pRF models that explained at least 15% of the variance.   The interaction between the presence of scotomas and the pRFs was modelled via a gain field (GF) model, Klein et al, 2014. A GF shifts the position and size of a pRF. The GF was centred at the edge of the scotoma. Its size was estimated by minimising the error between predicted and measured position shifts (SFR_scot vs SFR). Data was split in a training (50% of the data) and test set. Results: By comparing the pRFs estimated using SFR and SFR_scot we observed a change in position towards the scotoma, figure 1A. This effect was present throughout the six visual areas tested and it scales with visual hierarchy, figure 1B. The change in position is minimum near the edge of the scotoma and gradually increases with the distance from the edge, figure 1D. pRFs originally within the aSPZ shift radially towards the edge of the scotoma, see figure 1E. Regarding the pRF size, no significant changes were measured between SFR and SFR_scot, see figure 1C.   The common GF explained on average 65 % of the measured position changes. GF sizes tend to increase with visual hierarchy. Conclusions:   The presence of artificial scotomas resulted in pRFs shifts towards the scotoma’s edge throughout the visual cortex. We interpret this as an evidence of perceptual filling-in. The pRF shifts of neurons within the aSPZ resulted from an extrapolation process- it enables the stimulation by spared portions of the visual field. Surprisingly, the pRFs outside the aSPZ were also attracted towards the scotoma’s edge. This process was modeled using a GF. This suggests that attention directed to the edge of the scotomas modulates the pRFs position. Furthermore our results are in agreement with previous studies which suggested that filling-in depends on local processes generated at the edge of the scotoma in early visual areas, Komatsu, H. (2006). However in contrast with previous findings, we did not find evidence for expansion of the receptive fields. We conclude that, in response to an artificial scotoma – and most likely filling-in – there is a short-term reorganization not only in the aSPZ but throughout the visual cortex.

この発表は， フィリングインと呼ばれる面の形成に関する研究であり，私が用いているKanizsa figureでもフィリングインの要素が存在するため，非常に近い研究でした。また去年のOHBMで知ったpRFを用いており視覚的階層のどの領域でどこの補間を行っているのかを検討していました．このpRFは脳領域と視野領域の関連を見ることができる解析方法であることは理解していたが，其のために必要な設計やツールは理解できないままであった．しかし，MATLAB上で使用可能なツールボックスを教えていただくことができ，実験設計として注視点の色が変化することへの反応を含んでいたことから，実験設計に必要な要素を知ることができた．非常に興味深い解析方法なので調査し自身の研究にも使うことができるのであれば挑戦してみたいと思っている．
 

発表タイトル       ： Neural substrates of human facial emotion processing: evidence from an ALE meta-analysis 著者                  ： shaoling peng, Xinyu Liang, Chenxi Zhao, Gaolang Gong セッション名       ： Poster session Abstruct ： Background: There are six basic emotions: anger, fear, sad, happy, disgust and surprise[1]. To date, fMRI results regarding the human facial emotion processing for each of the basic emotions are inconsistent, possibly due to differences in subject group, stimulus materials, and experimental paradigms across studies. To address this, we here applied a coordinate-based activation likelihood estimation (ALE) meta-analysis [2, 3] to investigate neural substrates underlying human emotional face processing. Methods: Candidate articles were from 3 sources: 1) PubMed dataset; 2) studies listed by other related meta-analysis studies[4-6] and 3)references of studies retrieved by 1 and 2. Eligible studies were selected using following inclusion criteria: (1) fMRI or PET studies; (2) healthy subjects with age older than 18; (3) whole brain analysis; (4) emotional facial stimuli as experimental condition, with neutral face as control condition; (5) includes activating, rather than deactivating coordinates. Once the eligible studies were obtained, coordinates were extracted and analyzed using the ALE meta-analytic tools, where a cluster-level FWE threshold of p < 0.05 and a cluster-forming threshold of p < 0.001 were applied as the significant level. We first located brain areas that are activated under each basic emotion by analyzing studies with neutral faces as control stimuli. Next, studies of all basic emotions are combined to investigate emotional face processing in general. Results: 125 studies with a total of 2675 subjects were selected. The “surprise” condition was excluded, because of the very small number of eligible studies (i.e., 4). The ALE results for each basic emotion are shown in Figure 1. All basic emotions except “disgust” showed activation in the amygdala. Both “sad” and “happy” groups have only one significant cluster in the left amygdala, while “angry” and “fear” groups activated both the left and right amygdala. In addition to activation in the amygadala, watching “angry” faces also triggered activation in the right fusiform gyrus, while “fear” face processing triggered activation in the bilateral fusiform gyrus, right occipital lobe and left insula. In contrast, “Disgust” facial processing activated bilateral middle and inferior occipital gyrus. Comparing all emotional faces with neutral face, which putatively involves only the general emotional cognitive processing component, showed activation in the bilateral amygdala, bilateral fusiform gyrus, bilateral insula, bilateral thalamus, and bilateral middle and inferior occipital gyrus. Conclusions: Our results show shared and distinct patterns of activation when processing human faces with different basic emotions. As expected, the amygdala is pivotal to human emotional processing. Processing human emotional faces not only recruits brain regions involved in emotion processing, but also areas that are known to be engaged in face processing, e.g., the fusiform gyrus [7].

この発表は， メタアナリシスを用いた情動の研究でした．PubMedのデータセットを用いてメタアナリシスの1つであるALEを用いて情動に関わる脳賦活領域を抽出していました．125個の研究分，計2675人分のデータを用いてhappy，sad，disgust，angry，fearの各情動について検討を行っていました．また情動そのものに関連する脳領域の特定も行なっていました．参考文献を選択する基準として，fMRIまたはPET研究であること，18歳以上の健常人であること，全脳分析であること，実験条件としての感情的顔面刺激、対照条件として中立面を用いていること，座標が有効であることと設定していた．その結果，異なる基本的な感情を持つ人間の顔を認識する処理を行うときの活性化は異なるパターンを示しており，扁桃体が人間の感情処理にとって非常に重要であることがわかったと発表していました．メタアナリシスを調査していたときに，PICOという選択基準の存在を知ったが，実際にメタアナリシスを用いた研究がどのように論文を選択しているのかについてはあまり調査できていなかったので，今回この聴講を受けて理解が深まったと思います．
参考文献
1) OHBM2018 Annual meeting,
https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageID=3821
学会参加報告書

報告者氏名	古家知樹
発表論文タイトル	低周波振動振幅強度に基づく瞑想時脳状態の機能的脳分割
発表論文英タイトル	Functional brain parcellation of the meditative brain based on amplitude of low-frequency fluctuation
著者	古家知樹, 日和悟, 廣安知之
主催	Organization of Human Brain Mapping
講演会名	25th Annual Meeting of the Organization of Human Brain Mapping
会場	Auditorium Parco Della Musica
開催日程	2019/06/09-2019/06/13

 
 

1. 講演会の詳細

2019/06/09から2019/06/13にかけて，ローマのAuditorium Parco Della Musicaにて開催された25th Annual Meeting of the Organization of Human Brain Mappingに参加いたしました．この学会は，この学会は，Organization of Human Brain Mappingによって主催された国際学会で，ヒトの脳地図作成のための最新の研究について学ぶことを目的に開催されています．この分野の専門家とポスター発表を通じてディスカッションを行い，世界中の研究者と交流することが可能です．
私は全日程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，大塚，杉野，奥村（駿），山本，吉田，風呂谷，丹が参加しました．

2. 研究発表
   • 発表概要

私は11日に開催された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時間の発表時間となっておりました．
 
今回の発表は，Mapping the brain state behavior during meditation in low dimensional feature spaceです．以下に抄録を記載致します．

Introduction: Mindfulness meditation has the positive effects as it improves well-being by reducing stress and improves concentration. However, novices without meditation experience find it difficult to arrive at the meditation state. The purpose of this study is to examine the neural basis that makes meditation successful. The region of interest (ROI) used to determine the meditation state depends on brain region segmentation. Therefore, we propose a method for evaluating the meditation state based on brain activity intensity during meditation. In previous studies, the methods employed to divide brain regions were not based on brain activity. because it was thought that brain activity during meditation could not be evaluated correctly. However, in this study, characteristics of activity during meditation were examined based on brain segmentation defined by fractional amplitude of low frequency fluctuations (fALFF).   Methods: Fig.1 shows a brain segmentation method that quantitatively evaluates the participants meditation state based on fALFF (Zou et al 2008). Twenty-nine novices of meditation (22.9 ± 2.3 years; 6 females) and 7 practitioners of meditation (43.0 ± 9.1 years; 1 females) participated in this experiment. They performed 5-minute breath-counting meditation after 5-minute rest periods in an fMRI scanner. Brain activity of all participant s during rest and meditation was used for analysis. First, the z score (zfALFF) of fALFF, which is an indicator of local spontaneous brain activity in each voxel of 7 meditation practitioners, was calculated, and the brain region during meditation was segmented by simple linear iterative clustering (SLIC) (Achanta et al 2012). Furthermore, the correlation between the questionnaire and the area where fALFF increased was calculated for the 29 beginners. Finally, brain functions in the automated anatomical labeling (AAL) in regions with high correlation were examined.   Results: Fig.2 shows the results that division from the fALFF of the group of meditating practitioners. The number of divisions was 189 and it was shown that fALFF in 12 areas increased during meditation. The fALFF increased, there were left frontal sup (dlPFC), left Insula, left Cingulum Ant (ACC) and right Parietal Sup (SPL) all of which are areas of the task positive network (TPN) involved in meditation. In addition, Frontal Sup Medial (mPFC) and right Parietal Inf (IPL) were involved both of which are the areas of the default mode network (DMN) involved in mind wandering. Correlation between the questionnaire and the fALFF of the 29 beginners was calculated in the area where the fALFF increased. A significant correlation between fALFF and the questionnaire was shown in one area. Thus, participants who did not get distracted during meditation tended to show high activity intensity. Furthermore, a significant correlation was found in left SPL which is the region of the central executive network (CEN) involved in meditation. In addition, the left IPL and left precuneus were involved, which are areas of the DMN involved in mind wandering. From these results, it is suggested that participants with a high evaluation of the degree of realization of meditation have increased activity intensity of the TPN and the DMN during meditation. By changing segmentation based on the brain activity, the proposed method could determine brain-state during meditation. Thus, it is possible to determine the region of the brain involved during 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．本発表では質問者の名前は聞いていません．
 
・活動強度とはどういう指標なのか
この質問に対する私の回答は自発的な脳活動を示す指標で，ヒトは0.008から0.09Hzの周波数帯で自発的な脳活動が現れるとされており，その周波数帯での活動が全周波数帯にたいしてどのくらい含まれているかで算出されますと回答いたしました．
 
・SLICとは何か
この質問に対してSLICとは色や空間的特徴量を用いてクラスタリングする手法で，主に画像処理などで用いられており，k-meansを基に作成された手法であると回答致しました．
 
・パーセレーションは何を用いて行ったのか
この質問に対して，Pythonのscikit-learnを用いて行っていると回答いたしました．
Pythonにあまりなじみの無い方もおり，もっと詳しく説明する必要があったな反省いたしました．
 
・ネットワークは見ないのか
この質問に対してネットワークは協調の指標なので，まず活動が増加している領域が瞑想時の特徴を示していると考えるため本実験では活動をみていますと回答いたしました．
 
・この脳地図は何に良いのか
この質問に対する私の回答は今回の脳地図で実践者に共通して活動する領域を見つけることができたので，将来はこのような領域の用いて初心者の脳状態のフィードバックを行いたいと考えていると回答いたしました．
 
・パーセレーションする際の脳のテンプレートは何を使用しているのか
AALで領域が割り当てられている脳テンプレートを用いてパーセレーションを行っていると回答いたしました．
 
・fMRIのパラメータについて
　この質問はうまく聞き取れなかったのですが，TRに対するスライスの厚みが薄いというようなアドバイスと，やはり1.5Tではいまいちというようなご指摘をいただきました．
 
・ROIから何が言えるのか
　この領域はいずれ瞑想時の神経基盤の解明に役立つ領域であると回答いたしました．
 
・Dice係数とはなにか．どのように比較しているのか．
この質問に対する回答として，Dice係数はある領域のボクセルの空間的な重なりを測る指標です．と回答いたしました．どのように比較しているかというと，全領域のdice係数を被験者間で算出し，領域のdice係数が高い上位5%の領域を本研究のROIと定義しました．
 

• 感想

ポスターセッションの始めは聞きに来てもらえる人が少なかったのですが，時間が経つにつれて多くの人に来ていただきおもしろいと言っていただけました．自分の研究のアプローチ方法が新しく興味を持ってもらえる内容だと自信を得ることができました．しかし，英語での質問には内容を理解するまで時間がかかってしまい苦戦しました．ですが，自分の回答に納得していただける場面もあり、参加することができ良かったと感じています．同時にもっと高いレベルでディスカッションをするために勉強が必要であると強く感じました．
 

3. 聴講

今回の講演会では，下記の3件の発表を聴講しました．
 

発表タイトル       ：The effects of Vipassana meditation on brain network: a magnetoencephalography study. 著者                  ： Anna Lardone, Marianna Liparoti, Pierpaolo Sorrentino, Rosaria Rucco, Fabio Baselice, Francesca Jacini, Matteo Pesoli, Arianna Polverino, Roberta Minino, Emahnuel Troisi Lopez, Pietro Cipresso, Giuseppe Sorrentino, Laura Mandolesi セッション名       ： Posters Session Abstract            ： Introduction: In recent decades there has been growing attention to meditation. This interest is probably due to the increasing number of evidences showing how the practice of meditation may improve higher level of well-being and specific cognitive functions. Neuroimaging and electroencephalographic studies have shown that the brain connectivity of meditators changes as they meditate as well as in the resting-state. Furthermore, the constant practice of meditation causes widespread long-term changes in structural connectivity, suggesting that meditation might induce neural plasticity. Main areas involved are right orbito-frontal cortex, anterior cingulate cortex, right anterior insula, left inferior temporal gyrus, right hippocampus and amygdala (Newberg et al. 2014). The aim of our study is to compare the resting-state brain activity of mindfulness meditators with more than one year of experience and meditation-naïve controls recorded using a magnetoencephalography (MEG) in order to clarify the neural circuits that benefits of mindfulness meditation.   Methods: Twenty-six meditation practitioners, and twenty-nine controls matched for age, gender, education, race, and handedness were recruited. All participants were right-handed adults, had no general or mental illness, and were native Italian speakers. Meditators were trained to Vipassana Meditation (mindfulness meditation) and had an average of 6,41 (SE = 1,489) years of meditation experience. Both groups underwent five minutes of closed eyes resting-state MEG acquisition. The row data were cleaned from environmental noise, physiological and system related artifacts using Principal Component Analysis, Independent Component Analysis and visual inspection (Figure 1 a). Subsequently, the time series of neuronal activity were reconstructed in ninety regions of interests (ROIs) using the beamformer based on a template MRI (Figure 1 b) and then filtered in the classical frequency bands (delta, theta, alpha, beta, gamma) (Figure 1 c). To estimate the connectivity between ROIs, we calculated the Phase Lag Index (PLI) and the Minimum Spanning Tree (MST) was reconstructed (Figure 1 d & e) Finally, we compared topological metrics in meditators and non-meditators using permutation testing corrected for multiple comparisons through false discovery rate.   Results: Our findings reveal differences between the two groups in the resting-state condition. Compared to the non-meditator group, meditators show a higher degree (P = 0.009) in the right hippocampus in the theta band (Figure 2). The degree represents the number of connections incident upon a given node.   Conclusions: Previous studies have suggested that the right hippocampus is engaged during the creation of future events and in spatial memory processes (Bohbot et al. 1998). Furthermore, navigation abilities fully rely on the metric of theta oscillation (Buzsáki 2005). On this evidence the result of increased activation of the right hippocampus in theta band in meditators during resting state condition, supports the real possibility that meditation causes changes in brain networks and may improve specific cognitive functions, as spatial abilities and prospective memory. These changes could be seen in the perspective of treating degenerative pathologies characterized by alteration in the hippocampal areas and functional deficit in spatial orienting, such as Alzheimer disease.

この発表は，MEGを用いて瞑想時の神経基盤を解明するというものでした．データの次元削減方法としてPCAやICAを行い，Minimum Spanning Tree（MST）を用いて状態の推定を行っていました．MSTは以前ネットワーク勉強会でも取り上げた手法だったため，今後私たちの研究にも参考にできるように勉強したいと考えました．
 

発表タイトル       ：　Notizia dell’AFNI! AFNI now makes templates from your subjects easily! 著者                  ： John Lee, Paul Taylor, Robert Cox, Daniel Glen セッション名       ： Posters Session Abstract            ： Introduction: MRI and FMRI studies have standardized on just a few brain templates over the years. Commonly these have been limited to just the well known individual template, the N27 brain template from 27 scans of a single individual, or one of several MNI human brain templates, such as the 2009 version made by combining 152 (mostly young adult) subjects. Additional contributions have occasionally been made for other age groups, like pediatric templates[1], or other species, like the macaque with similar individual (D99)[2] and group-based templates (NMT)[3]. Templates serve two purposes-correspondence+reference. Group templates have an advantage over a selected individual template because idiosyncratic structures are avoided, and common features are extracted. Earlier group templates were limited by their alignment algorithms to simple affine transformations, and a mean computed across subjects. The result was a blurry combination retaining only the most major features of a brain. Recent implementations make use of nonlinear warping that dramatically improves feature correspondence across subjects. Here we introduce a new AFNI tool, make_template_dask.py, for the creation of group-based templates, across a desired group or species. Importantly, the program uses Dask [4] to efficiently run this intensive, multilevel processing efficiently on any computer setup (single CPU, multicore or cluster). Much of the processing repeats similar processing for every subject in the group, and, consequently, is “embarrassingly parallel”. The script makes use of existing AFNI tools and wraps them in the Dask parallelization toolbox.   Methods: The script does a series of steps for each subject and then for the means across subjects:Unifize (i.e., bias correct) subjects to standardize intensities Skullstrip Align centers to some starting base template or subject Rigidly align (rigid component of an affine alignment) to the initial template Compute mean across all rigidly aligned subjects -> rigid template Affine align all rigid align subjects to rigid template Compute mean across all affinely aligned subjects -> affine template Nonlinearly align all affinely aligned subjects to affine template in large voxel neighborhood (patch size 101mm) Compute mean across all nonlinearly aligned subjects -> nl0 template Repeat nonlinear alignment and mean computation of two previous steps four more times with successively finer voxel patch sizes (49, 23, 13, 9 mm). -> nl1,2,3,4 templates Dask provides a convenient and scalable framework for using Python scripts, parallelizing the processing into multiple threads to be executed on a variety of platforms. make_template_dask.py is written to be run in parallel on a server or desktop with many CPUs or on a computing cluster like the NIH Biowulf SLURM cluster [5] While Dask simplifies this task, there are still inherent issues regarding individual commands that are written to use specific filenames with each instance. This problem was solved with parallelization by subject and processing each subject’s data in a separate directory, until individual results are brought together for each averaging step. There are additional options for anisotropic smoothing,unifizing of the nonlinear mean results, and for skipping any of the steps. There are further options to allow for restarting the script with already existing datasets in case of server or cluster failures or timeouts.   Results: Results will be shown for test group of 15 subjects from OpenFMRI, Indian brain template groups, Toddler age group with example workflow plots.   Conclusions: make_template_dask.py simplifies the process of making group templates with a command line interface with an efficient use of compute services, largely independent of number of subjects. This script can be applied to a variety of subject groups and compute platforms. New templates can be created for groups based on age, geography, species… with the goal of making these templates more relevant to the research study.

この発表はfMRI研究で用いる新しい脳テンプレートを作成した研究でした．これまで多くの研究で用いられているMNIテンプレートなどは少人数かつ年齢層を偏ったものが多く，この研究は子供の脳から一般的に使用可能なテンプレートを作成していました．内容も興味深かかったのですが，大きく書かれた”One(brain) for All, All(brains) for One ”というキャッチ―なフレーズが目を惹きました．まずは人に注目してもらえるようなポスター作成の重要性を学び今後ポスターを作成する際には参考にしたいと考えました．
 

発表タイトル       ：　Assessing multisite reproducibility of parcellation methods using traveling subjects 著者                  ： Giuseppe Lisi, Ayumu Yamashita, Noriaki Yahata, Takashi Itahashi, Takashi Yamada, Naho Ichikawa, Masahiro Takamura, Yujiro Yoshihara, Akira Kunimatsu, Naohiro Okada, Hirotaka Yamagata, Koji Matsuo, Ryu-ichiro Hashimoto, Go Okada, Yuki Sakai, Jin Narumoto, Yasuhiro Shimada, Kiyoto Kasai, Nobumasa Kato, Hidehiko Takahashi, Yasumasa Okamoto, Saori Tanaka, Okito Yamashita, Hiroshi Imamizu, Mitsuo Kawato, Jun Morimoto セッション名       ： Posters Session Abstract            ： Introduction: When collecting large neuroimaging data associated with brain disorders, images must be acquired from multiple sites (e.g. hospitals) because of the limited capacity of a single site. Site differences represent a great barrier, making data harmonization necessary (Yamashita et al. 2018). This is especially true when considering the problem of building a connectivity-based classifier that generalizes from one site to another (Yahata et al. 2016). In this context, brain parcellation is a critical step, since brain regions computed (e.g. by independent component analysis, ICA) on data from site A, may not generalize well to data from site B. Previous works have systematically analyzed the impact of different parcellation methods, by comparing either classification accuracy (Dadi et al 2018), or reproducibility of parcellations across scans (using the same scanner) within a subject (Arslan et al. 2018). In this study, we take one step further, by investigating the reproducibility of common parcellation methods across sites using a travelling subject dataset. Methods: We use the travelling-subject dataset collected in Yamashita et al. 2018. Nine healthy participants were scanned at each of 12 different sites in Japan, 3-4 times each, producing a total of 411 scan sessions. Data were preprocessed according to the procedure described in Yamashita et al. 2018. We take into consideration all the parcellation methods in Dadi et al. 2018, excluding the structural and functional pre-computed atlases. As a result, we include in the analysis two linear decomposition methods (i.e. Canonical ICA and Dictionary Learning) and two clustering methods (K-means and Ward clustering) (Abraham et al. 2014). The number of parcellations is set to 100 for every method, as Dadi et al. 2018 found it to be a sufficient number for good prediction. Then regions are extracted by breaking out clusters in their connected components. During this procedure, we remove spurious regions of size < 1500mm3 (Dadi et al. 2018).We apply the above four methods to the whole group of subjects, separately for each site, computing a total of 48 (4 methods x 12 sites) different parcellations. Separately for each parcellation method, we compare the reproducibility across sites using parcellation similarity scores (Dice Similarity and Joined Dice Similarity, Arslan et al. 2018) for each pairwise combination of sites.   Results: The Dice Similarity scores (Fig. 1a) show that CanICA produces the most reproducible parcellations across sites, followed by Dictionary Learning, Ward Clustering and k-means clustering. When using the Joined Dice Similarity score (Fig. 1b), Dictionary Learning gets closer to CanICA. These quantitative results are confirmed by visual inspection. For example, Fig. 2 shows a better similarity across sites in the regions extracted by CanICA, compared to k-means.   Conclusions: Linear decomposition methods provide better reproducibility scores. These methods also achieved the best classification performance on the benchmarking study by Dadi et al. 2018. On the other hand, Ward and k-means clustering show relatively poor reproducibility scores, which is in agreement with Arslan et al. 2018. As suggested in Dadi et al. 2018, the linear decomposition methods provide soft assignments that capture uncertainty, allowing to define overlapping regions. It should be noted that the dice scores of our dataset are generally lower than those reported in Arslan et al. 2018, suggesting that scanner differences in the travelling subject dataset add uncertainty that should be taken into account in order to build truly generalizable classifiers. In future, we will analyze additional measures such as homogeneity (i.e. correlation among the vertices of a region), that would allow the inclusion of pre-computed atlases in the analysis. This study provides the basis for building parcellation methods that are robust against site differences.

この発表では，最良な脳分割手法についての研究が行われており，パーセレーションはK-meansやcanicaなどの手法を用いて行っていました．K-meansはいまいちという結果が示されていたことから，k-meansを基にしているSLICについてもう少し理解を深める必要があると感じました．アトラス間類似度の評価方法として，dice係数のほかにもさまざまな指標を用いて行っており，参考になるものが多いと感じました．
学会参加報告書

報告者氏名	風呂谷侑希
発表論文タイトル	低ランク近似クラスタリング手法を用いた機能的ネットワーク構造の推定
発表論文英タイトル	Extracting functional network structures using low-rank matrix factorization-based matrix clustering
著者	風呂谷侑希, 日和悟，谷岡健資，宿久洋，廣安知之
主催	Organization for human brain mapping
講演会名	25th Annual Meeting of the Organization of Human Brain Mapping
会場	auditorium parco della musica
開催日程	2019/06/09-2019/06/13

 
 

1. 講演会の詳細

2019/6/9から2019/6/13にかけて，イタリアのauditorium parco della musicaにて開催されました25th Annual Meeting of the Organization of Human Brain Mappingに参加いたしました．2019 OHBM annual meetingは，Organization for human brain mappingによって主催された国際会議で，ヒトの脳組織および脳機能のマッピング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．私は全日程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の学生として古家さん，大塚さん，奥村（駿）さん，杉野さん，山本さん，吉田さんM1の学生として丹が参加しました．
 

2. 研究発表
   • 発表概要

私は10日から13日の4日間開催されたPoster Sessionのうち，12日のPoster Sessionに参加いたしました． 発表形式はポスター発表で，1時間の発表時間となっておりました．
今回の発表は，Extracting functional network structures using low-rank matrix factorization-based matrix clusteringです．以下に抄録を記載致します．

Background: Working memory (WM), which is a temporary storage system that processes information, is necessary for daily life (A.D. Baddeley 2000). In the present study, the influence of the WM load on the changes in functional connectivity (FC) networks was examined. Given the high-dimensionality of the brain function data, it is difficult to find a characteristic pattern among participants using such as the arithmetic average because the variability of the data is high. In contrast, an analysis involving dimensionality reduction is necessary to determine latent feature patterns. Therefore, a method for extracting a characteristic function network structure which changes depending on the WM load based on low-rank factorization-based matrix clustering (D.J. Simon and J. Abell 2010) was proposed and applied to fMRI data measured during an N-back task.   Methods: A total of 29 healthy adults (22.3 ± 0.17 years; 10 females, 19 males) performed N-back (N = 1, 2, 3) tasks in an fMRI scanner. Furthermore, the FC matrix between the brain areas was calculated for all the 116 regions defined by the automated anatomical labeling (AAL). As illustrated in Fig.1, a method was proposed to estimate a single FC matrix expressing the brain functional network characteristic of plural FC matrices under the same experimental condition through low-rank matrix factorization-based matrix clustering. The squared error between the deriving matrix and the input matrices was minimized using a numerical optimization algorithm. The method enabled the extraction of the latent features from the data in a high dimensional space. Additionally, our method performed clustering of the brain regions in the derivation process of the matrix factorization so as to extract the module structure existing in the network represented by the derived low-rank matrix.   Results: The proposed method was performed on the FC matrix at each N-back load of 29 participants, while a single representative FC matrix associated with each load was estimated. Fig. 2 indicates the brain function network structure estimated for the 1-, 2-, and 3-back tasks. Specifically, the FC of the prefrontal cortex was confirmed in both the 1- and 2-back tasks, whereas other brain areas involved in the WM activity, including the left superior frontal gyrus (SFG), left middle frontal gyrus (MFG), the left/right medial frontal gyrus (SFGmed), functioned as the same module in the 1-back task, and FC that formed a single module in the area of somatosensory cortex was also confirmed. In contrast, in addition to the right SFG, left/right MFG, the FC of the left/right opecular part of inferior frontal gyrus (IFGoper) and left/right triangular part of inferior frontal gyrus (IFGtriang) belonging to the Broca’s area was also confirmed in the 2-back task. Moreover, the left/right inferior parietal lobule (IPL) and the left/right supramarginal gyrus (SMG) were confirmed to present the same module to the prefrontal cortex (PFC). Finally, the FC of the brain regions mainly related to the visual cortex (VC) and cerebellum was confirmed in the 3-back task, while that of the PFC was not estimated. The regions belonging to multiple clusters were derived for all WM loads and were considered to represent hubs connecting the different modules.   Conclusions: In the present study, the FC matrix of 29 healthy adults during the N-back tasks was measured using fMRI. Additionally, the single representative FC matrix was estimated for each N-back load from the FC matrices of the multiple participants using the proposed method. The results suggested that the WM load-dependent changes in a functional network and its module structure could be identified by our proposed method.

 

• 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．本発表では質問者の名前は聞いていません．
 
・connectivity算出における窓関数はどれくらいなのか
この質問に対する私の回答として，「本研究では，ダイナミック解析は行っておらず，connectivity用に計測した250秒の血流変化量の時系列相関を算出している」と説明致しました．
 
・それぞれの被験者データから集団のモジュール構造を推定する意味は
この質問に対する私の回答として，「集団間のモジュール構造を推定することでワーキングメモリにおける脳機能ネットワークを解釈しやすい形で表現することができる」と回答しました．
 
・設計変数の実行可能領域は何なのか
この質問に対して私はその場でお答えすることができませんでした．ポスター発表終了後考えたところ，実行可能領域は設計変数 p=脳領域数，c=クラスタ数と説明する必要があると感じました．
 
・この提案手法はソフトウェアを用いているのか自作のコードなのか
この質問に対しての私の回答として，「共同研究者の方々と組んだ自作のソースコードである」と回答いたしました．
 
・前頭前野の部分が推定されていないのはなぜか
この質問に対する回答ですが「私の手法はモジュール構造を強調する手法であるため，課題時の前頭前野の機能はモジュール構造を有していない可能性がある．」と回答いたしました．
 
・1-backと2-backで出ているクラスタは同じクラスタなのか
この質問に対する私の回答として，「同じようなクラスタも存在するが，異なるクラスタ構造も存在する」と回答いたしました．また補足として，「図のノードの色は同じクラスタを表しており，ノードの大きさは何個のクラスタに属するか表現していると説明し，1-back，2-backのノードの色に依存関係はない」と説明しました．
 
 

• 感想

今回の学会が，初の学会参加かつ初の国際学会であり，とても緊張しました．また英語でのポスター発表であり，英語で研究内容を伝えることの難しさを痛感いたしました．しかし，研究内容に興味を持って頂けた方々とのコミュニケーションを通して，英語で説明することへの抵抗がなくなり，より英語がうまく話せるようになりたいというモチベーションに変わりました．今回，質問があまり聞き取れず，聞き直すことが多く，また原稿に助け借りることが多かったので，今後は英語の勉強により一層取り組んでいきたいと思います．本学会を通して，慣れない異国の地で様々な文化や価値観に触れることで，自分のレベルアップのためにすべきことや日本や海外ぞれぞれの良し悪しなどが分かり，グローバルな視野が少しは付いたと感じとても素晴らしい経験になりました．
 

3. 聴講

今回の講演会では，下記の3件の発表を聴講しました．
 

発表タイトル       ：　Persistent hippocampal neural firing and hippocampal cortical coupling predict working memory load 著者                  ： Ece Boran, Tommaso Fedele, Johannes Sarnthein セッション名       ： Poster Session Abstruct            ：   Introductions: Hippocampal activity is known for its role in cognitive tasks involving episodic memory or spatial navigation, but its role in working memory and its sensitivity to workload is still under debate. The maintenance of items in working memory relies on persistent neural activity in a widespread network of brain areas. It remains, however, unclear how load on working memory influences persistent neural activity, particularly in the hippocampus and in its functional connections.   Methods: Here, we investigated hippocampal activity while subjects maintained sets of letters in verbal working memory for a few seconds to guide action. We used the Sternberg paradigm as a standard tool to study working memory. Subjects were presented with sets of 4,6 or 8 letters for 2 seconds (encoding), maintained them in working memory for 3 seconds (maintenance), and – upon viewing a probe letter – responded by pressing “in” or “out” (retrieval). In nine patients with epilepsy, we recorded single neuron firing and intracranial EEG in the medial temporal lobe and EEG from the scalp. We used machine learning tools to predict the subject’s behavior on a trial-by-trial basis. We calculated the phase locking index over time and spectral frequency to characterize the functional connectivity between hippocampal EEG and scalp EEG.   Results: Hippocampal neural firing distinguished between the periods of a trial and increased with the number of letters to be memorized (Figure 1). Along the periods of a trial, hippocampal firing differentiated between success and error trials during stimulus encoding, predicted workload during memory maintenance, and predicted the subjects’ behavior during retrieval. The hippocampal neurons did not encode for individual items of memory content, but rather the number of letters maintained in memory. This property was known from frontal cortex and is shown here for neurons within the medial temporal lobe for the first time. After having demonstrated that the hippocampus is involved in the processing of verbal working memory load at the single neuron level, we studied how the hippocampus is embedded in the working memory network. We found that hippocampal oscillations coupled to scalp EEG in the theta-alpha range, exclusively during maintenance, and with increasing workload (Figure 2). This demonstrates a network for working memory that is bound by coherent oscillations involving cortical areas and persistent hippocampal neuron firing.   Conclusions: We are the first to combine hippocampal single-unit activity with scalp-recorded EEG, all in a task that is feasible only for human subjects. Connecting the single-neuron scale to the cortical population scale represents a breakthrough in the study of human functional neuroanatomy. These findings allowed us to clarify the role of the hippocampus with single units and characterize its functional connectivity in the working memory network.

この発表はワーキングメモリ課題おける海馬の機能的結合に注目した研究でした．海馬の活動は，エピソード記憶を含む認知課題の役割として知られていますが，ワーキングメモリにおける役割および負荷に対する変化はいまだに議論中であり，本研究の結果としてはワーキングメモリに与える負荷量の増加に伴って海馬の神経発火も増加するという結果でした．この研究結果の内容は自分の研究と照らし合わせて考察する必要があると感じました．また，この研究ではワーキングメモリを研究するための基本的なツールとしてSternberg paradigmというものを使っていて調査する必要があると感じました．
 

発表タイトル       ：Modeling and Analysis Methods – Uni/multi-variate analysis 著者                  ：Cedric Huchuan Xia, Zongming Ma, Danielle Bassett, Theodore Satterthwiate, Russell Shinohara, Daniela Witten セッション名       ： ORAL SESSION: Modeling and Analysis Method – Uni/multi-variate analysis Abstruct            ：   Introductions: Relating connectomic features to subject-level measures is a popular approach to study the brain-behavior relationships. However, most studies have relied on mass univariate methods, which are vulnerable to Type II errors due to extensive comparisons that must be accounted for. On the other hand, studies using common multivariate analyses often ignore the innate structure of brain network data, producing results that are difficult to interpret. Here, we introduce a new penalized regression method, specifically designed to analyze the relationship between high-dimensional connectomic data and covariates of interest.   Methods: Given n subjects, let Ai denote the connectivity matrix for the i-th subject, where p is the number of nodes. The p nodes can be sorted into K communities. For each subject, q covariates have been measured, so that Xfi is the f-th covariate measurement for the i-th subject (Fig 1a). We consider the model Ai = Θ + Σqf=1Xfi•(WΓfWT), where the mean connectivity matrix Θpxp captures the edge-level information, WpxK indicates the community membership for each node, and ΓqxKxK captures the community-level information for each covariate. We make two assumptions: 1)Θ is low-rank, i.e. it can be represented by a low-dimensional subspace, and 2)Γ is sparse, i.e. some of the elements are exactly zero. They can be implemented via a nuclear norm (λ1) and a L1 norm (λ2) penalty, respectively. Finally, the convex optimization problem can be solved by a block coordinate descent algorithm. As this method is designed to incorporate node-, edge-, and community-level information, we refer to it as multi-scale network regression (MSNR). We exercised MSNR by applying it to data from the Philadelphia Neurodevelopmental Cohort. We analyzed 1015 subjects (age 8-22), who completed a resting-state fMRI acquisition, had adequate data quality, and were not excluded for medical co-morbidity. We used a validated preprocessing pipeline to minimize motion artifact. We constructed functional connectivity matrices using a commonly-used parcellation. To evaluate the performance of all models, we randomly selected 10% of the data as a validation set. In the remaining 90% of the data, we tuned λ1 and λ2 via a 9-fold cross-validation (Fig 1b). We evaluated the final model in three approaches: 1) out-of-sample prediction, 2) permutation testing, and 3) bootstrapping (Fig 1c). Finally, we also benchmarked the MSNR’s predictability and interpretability against other common methods, such as linear models using community-mean and individual edges as features (Fig 2a).   Results: We successfully trained a multi-scale network regression model linking high-dimensional functional connectivity to three covariates: motion, age, and sex. MSNR performed well in the validation set, with a prediction error comparable to the mean error in the training set (Fig 2b). Notably, its performance was also significantly better than the null distribution generated by permuted data (p < 0.001). In contrast, the community-based model generalized poorly to the validation set, whereas the edge-based model was on par with MSNR. However, both common methods had substantially less interpretable coefficients than MSNR (Fig 2c, d). Finally, MSNR revealed known connectivity-covariate relationships, such as distance-dependent motion artifact, developmental effects, and sex differences (Fig 2e).   Conclusions: By taking into account of node-, edge-, and community-level information, we developed a multi-scale network regression approach that achieves a balance between prediction and interpretability. Empirically, we demonstrated its advantages over traditional methods and its ability to uncover meaningful relationships. This new method could prove useful in future studies of brain-behavior relationships.

この発表で着目したのは自分と同じように多変量解析におけるモデルや新たな解析手法を紹介しているところです．一般的な多変量解析を用いた研究では，脳機能ネットワークの本質的な構造を無視することが多く，解釈が困難な結果が生じる場合があります．本研究では，その解決方法として特に高次元のconnectomicデータと関心のある共変量間の関係を分析するために設計された，新たな制約付き回帰法を紹介しており，解析手法が私たちの研究にも参考にできると感じました．
 

発表タイトル       ：Community organization of brain networks and hub disruptions in chronic pain 著者                  ：Camille Fauchon, David Meunier, Kasey Hemington, Joshua Cheng, Rachael Bosma, Natalie Osborne, Anton Rogachov, Andrew Kim, Robert Inman, Karen Davis セッション名       ： Poster SESSION Abstruct            ：   Introductions: We previously reported that patients with chronic pain exhibit abnormalities of within- and cross-network functional connectivity, and regional dynamics within the dynamic pain connectome [1-5]. However, the extent and nature of changes in brain networks topology is poorly understood. Thus, the aim of this study was to model the brain as a modular network [6] using community network graph analysis based on resting-state fMRI (rsfMRI) data in patients with chronic low back pain.   Methods: Forty-five right-handed male chronic pain patients with ankylosing spondylitis (AS) and age/sex-matched healthy controls (HC) were recruited and provided informed consent to the study approved by our local research ethics board. AS predominantly occurs in young men with relatively few co-morbidities. Inclusion criteria for the patients were 18-65 years old, pain for >6 months, stable medications, and absence of other major diseases. Each subject had a 3T MRI session to acquire high resolution T1 anatomical and 10-minute rsfMRI scans. Data were preprocessed using FSL and parcelled based on the HCP atlas [7] using the open-source python package Nipype [8] to create 180 anatomical regions (nodes) in each hemisphere. For each subject, the raw time-series data were averaged over the voxels within the HCP areas and extracted using the graphpype functions of the neuropypcon package. Individual functional connectivity matrices were generated by computing the Pearson correlations between every pair of nodes (360 x 360). A network density threshold of 10% was applied to remove weak correlations. We then conducted a modular analysis using Radatools software [9] as previously described [6] based on the average Z-correlation matrix for each group. This segregated the network into modules of regions (i.e., communities) that work tightly together as a function of their level of mutual inter-connectivity [6]. We computed the networks metrics to characterize regions which acted as an inter-modular hub (e.g., betweenness centrality), or as an intra-module hub (e.g., degree Z-score). Permutation tests (n=500) were applied at the level of nodes and subjects to determine statistical significance of a network (p< 0.05; corrected for multiple comparisons).   Results: In both the HC and AS groups, there was a significant network structure of six connected communities compared to a random network, with similar modularity and no inter-hemispheric differences: 1) fronto-parietal regions (i.e., precuneus, medial and dorsolateral prefrontal cortex) including areas of the default mode network, 2) regions of the occipital cortex (i.e., calcarine, cuneus) of the visual network, 3) sensorimotor and salience areas including the insula, and mid/anterior cingulate cortex, 4) parietal sensorimotor areas (i.e., SMA, S1, M1), 5) temporal cortices and subcortical regions, and 6) orbito- and ventro-medial prefrontal cortex (vmPFC). Furthermore, the chronic pain AS group showed inter- and intra-modular hub disruption compared to the HC group as follows: 1) low betweenness centrality in the rostral anterior cingulate cortex and the anterior intra-parietal area, 2) high betweenness centrality in the precuneus/posterior cingulate cortex (PCC), the anterior mid cingulate cortex, and the vmPFC, and 3) high intra-modular degree in the inferior parietal area, PCC, posterior mid-cingulate cortex, and visual cortex.   Conclusions: Chronic pain patients had a normal number of communities but with abnormal membership and hub disruptions. These results provide evidence of brain network reorganization in chronic pain, and provide a framework to study the effects of pain on the brain from a network perspective. Our future analysis will assess the impact of individual differences (e.g., sex differences) in brain network topology and its interaction with chronic pain.

この発表で着目したのは脳機能ネットワークのモジュール構造をモデル化していることです．本研究では従来とは異なり，Radatoolsというソフトウェアを使用してモジュール構造の分析を行っており，またハブとして機能している領域を特徴付けているのが自身の研究と近く私の研究の参考になると感じました．データとしては慢性疼痛の患者のデータを使っており，結果として，正常者とモジュールの数は変わらないがハブの領域やモジュール内の構造が変わるという結果でした．
学会参加報告書

報告者氏名	吉田早織
発表論文タイトル	ダーツ投てき時の脳活動領域のためのfNIRSの計測データの体動除去手法の確立
発表論文英タイトル	Motion artifacts removal method for fNIRS data to examine brain activity during dart throwing
著者	吉田早織, 日和悟, 竹田正樹，廣安知之
主催	Organization for human brain mapping
講演会名	OHBM2019
会場	Auditorium Parco Della Musica / Rome
開催日程	2019/06/9-2019/06/13

 
 

1. 講演会の詳細

2019/06/9から2018/06/13にかけて，イタリアのローマにて開催されました2019 annual meetingに参加しました．この学会は，Organization for human brain mappingによって主催された国際会議で，ヒトの脳組織および脳機能のマッピングに関する研究に携わる様々な背景を持つ研究者を集め，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^（１．本研究室からは他に，廣安先生，日和先生，M2の学生として古家，大塚，奥村（駿），山本，杉野，M1の学生として風呂谷，丹が参加しました．
 

2. 研究発表
   • 発表概要

私は13日のポスターセッションおよびポスターレセプ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，計3時間参加者の方と議論を行いまいした．
今回の発表は，「Motion artifacts removal method for fNIRS data to examine brain activity during dart throwing」について発表いたしました．以下に抄録を記載致します．

Introduction Habitual darts training is said to have the potential to improve cognitive function (Takeda et al. 2017). Furthermore, using the cognitive function test, it has been shown to increase the short-term memory ability. It is necessary to examine this from the aspect of brain function. In this study, we measured brain activity in participants while throwing darts, using functional near-infrared spectroscopy (fNIRS). However, the measurement data include many motion artifacts (MAs) due to the throwing motion. Although several types of MA removal methods have been developed, their effectiveness for removing MAs due to dart throwing has not yet been investigated. Here, we performed removal of MAs in dart throwing using multiple methods and compared the results.   Methods Twelve healthy adults participated in the experiment. We measured measure oxy- and de-oxy hemoglobin (Hb) concentration changes during dart throwing using LABNIRS (Shimadzu, 38 CH, 37 Hz). Each subject repeatedly tried to throw darts 9 times in total. A voice instruction system told the subject when to begin the action. The oxy-Hb data were analyzed using Homer2 (Huppert et al. 2009) NIRS processing package functions in MATLAB (Mathworks, MA USA). The four different MA removal methods, targeted principal component analysis (tPCA), principal component analysis (PCA), MA reduction algorithm (MARA), and kurotsis wavelet (kWavelet) were applied to the oxy-Hb signals measured. Then, they were bandpass filtered (0.01 – 0.5 Hz). The function of hmrMotionArtifact implemented in Homer2 was used to identify the sections with MAs. Three sets of control parameters for the MA removal methods used in previous studies were set and compared each other. Additionally, we examined the differences in activated regions between the methods using general liner model (GLM).   Results After applying the MA removal methods, the section specified as the MA component was no longer observed. The number of the identified MA components differed depending on the parameter value and type of the MA removal methods (Fig. 1). The GLM analysis was applied to the oxy-Hb data after the MA removal (Figure 2). No regions were activated for data to which PCA was applied. The results of the data to which tPCA was applied varied depending on the parameter value. The results of MARA and kWavelet were similar to those only bandpass filtered.   Conclusion Oxy-Hb changes during dart throwing were measured using fNIRS. The measured data included MAs due to dart throwing. Several removal methods were applied and their results were compared each other. We have revealed that the result of each method differed from other methods and highly depends on its parameter setting. It is necessary to optimize it for the experiments.

• 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．質問者の氏名を控え損ねてしまいました．
 
・質問内容1
質問は，LABNIRSのデータをHomer2のデータ形式に変換するのが難しかったのではないかというものでした．この質問に対してはShimadzu2nirsを使用すると容易でしたと答えました．
 
・質問内容2
質問は，チャンネル配置を作るのはどうしたのかというものでした．この質問に対しては，基準点の値の微調整が難しかったと答えました．
 
・質問内容3
質問は，どのMA除去手法が有効であったのかでした．この質問に対しては，ウェーブレットフィルタでしたと答えました．
 
・質問内容4
質問は，ウェーブレットフィルタを行う際のiqrの設定はいくつにしたのかというものでした．この質問に対しては，1.5に設定しましたと答えました．
 
・質問内容5
質問は,  fNIRSとfMRIの違いについてでした．この質問に対しては，fNIRSは自然な状態で計測可能であるため，ダーツ投てき時の脳活動を計測することが可能ですと答えました．
 
・質問内容6
質問は，fMRIでは数mmでも動くとモーションアーチファクだがfNIRSではどれぐらい動くと影響するのかというものでした．この質問に対しては，何mmという値ではわからないと答えました．
 
・質問内容7
質問は，motion artifact detection algorithmのパラメータは自分の計測データに合わせるべきであるというものでした．この質問に対しては，今回は論文を参考にしましたが今後パラメータを検討したいと思いますと答えました．
 
・質問内容8
質問は，これからどうするのかというものでした．この質問に対しては，被験者数を増やしますと答えました．

• 感想

今回，私は初めての学会参加でした．準備が間に合わず，先生の力を借りることになり出発前から反省することばかりでした．また，発表時は方法や結果から尋ねられることが多く，最初は慌ててしまいましたが，発表しているうちに落ち着いて答えることができたように思います．しかし，英語を聞き取る能力や伝える能力が乏しく，議論が進まなかったことがあったのが悔しかったです．今回の学会では，準備不足や英語力の不十分さなど反省することが多くあったため，今後に活かしていきたいと思います．
 

3. 聴講

今回の講演会では，下記の4件の発表を聴講しました．
 

発表タイトル       ： A clinical fMRI protocol for cognitive-motor dual task 著者                  ： Oana Rus-Oswald1,2, Julia Reinhardt1,3, Céline Bürki2, Stephanie Bridenbaugh2, Christoph Stippich1, Reto Kressig2, Maria Blatow1 セッション名       ： Poster Session Abstruct            ： Introduction: Gait analysis involving cognitive-motor dual task (DT) is used as a diagnostic tool in geriatric populations (e.g. “walking while talking”)[1]. Cognitive-motor interference effects, measured as decrease of walking speed and increase of step variability during DT as well as decreases in cognitive performance have a high predictive value for future fall risk and cognitive decline[5]. In our previous study, we demonstrated the feasibility of performing the cognitive-motor DT in the fMRI environment using an MRI-compatible stepping device and evaluated the neural correlates of the DT costs [2]. In the present study, we aimed to optimize the DT fMRI protocol with respect to task difficulty, duration and signal robustness in order to be able to apply this fMRI protocol in clinical context. Furthermore, we compared the stepping DT paradigm to a finger tapping DT paradigm, to evaluate if the latter brings similar results in terms of DT difficulty and signal robustness.   Methods: 30 elderly healthy subjects (mean age ± SD: 70.2 ± 4.97) participated in the fMRI study which included the performance of a cognitive task (verbal fluency and serial subtraction) and a motor single task (ST; stepping or finger tapping) and the combination of both, i.e. a cognitive-motor DT. Data analysis was performed using standardized routines in BrainVoyager. First, the group level analysis based on the contrast task vs. baseline, was performed using a separate subjects fixed effects analysis. Second, a region-of-interest analysis (ROI) at the individual level of each subject was performed, employing a dynamic threshold technique[3, 4]. Further, the ROI based DT costs were computed based on the individual difference of activation between ST and DT.   Results: During cognitive-motor DT the primary and secondary motor as well as parietal and prefrontal areas were active at group level. Activation of motor areas was decreased in DT as compared to the motor ST, according to our previous findings[3]. Activation of parietal and prefrontal areas was on average equivalent or increased in DT as compared to ST. The stepping DT paradigm was more distinctive (higher ROI occurence and activation strength on individual level) between ST and DT than the finger tapping DT paradigm. At the individual level the following ROIs showed robust activations in terms of occurence probability and signal strength, measured in the left hemisphere: primary motor cortex (M1), supplementary motor area (SMA) and superior parietal lobule/intraparietal sulcus (SPL/IPS). The neural correlates of DT costs computed in the stepping condition in SPL/IPS enabled to descriptively separate the subjects into two groups, one with high and one with low DT costs based on their individual activation differences.   Conclusions: With this study, we propose an optimized cognitive-motor DT fMRI protocol and a standardized individual analysis routine to measure the neural correlates of cognitive-motor interference effects during DT. In the future this fMRI protocol may be evaluated in the clinical setting, i.e. in patients with mild cognitive impairment and might enable the early detection of motor and cognitive decline based on the obtained neurofunctional markers and preferably before the structural degeneration process occurs.

この発表は，fMRIを用いて運動タスクと認知タスクのデュアルタスク時の脳活動領域について検討されたものでした．運動タスクには足のステップ，認知タスクには引き算または言語記憶が用いられていました．本研究では，高齢者30名を計測しており，脳領域はPFC，M1，SMA，SPL，IPSに注目していた．結果ではM1は全被験者デュアルタスク時のほうがシングルタスク時より脳活動が低下していたが，SPLは半分の被験者はM1と同様に脳活動が低下していたが，半分の被験者では脳活動が増加していたことが示されていた．デュアルタスク時の高齢者の脳活動を計測することは，私の研究の最終目標と一致しており，注目する脳領域などを参考にしたいと思いました．
 

発表タイトル       ：　Motor imagery and visual neurofeedback to activate the swallowing network 著者                     ： Guilherme Wood1, Doris Grössinger1, Silvia Kober1 セッション名       ： Poster Session Abstruct ： Introduction: Motor imagery of movements is used as mental strategy in neurofeedback applications to gain voluntary control over activity in motor areas of the brain. In the present fMRI study, we first addressed the question whether motor imagery and execution of swallowing activate comparable brain areas, which has been already proven for hand and foot movements. Prior near-infrared spectroscopy (NIRS) studies provide evidence that this is the case in the outer layer of the cortex. With the present fMRI study, we want to expand these prior NIRS findings to the whole brain. Second, we used motor imagery of swallowing as mental strategy during visual neurofeedback to investigate whether one can learn to modulate voluntarily activity in brain regions, which are associated with active swallowing, using real-time fMRI.   Methods: Eleven right-handed, healthy young adults (4 male, 7 female, mean age = 29.18 years, SD = 5.62) took part in this study. All participants gave written informed consent. During the first session, participants performed the functional localizer task to identify brain areas, which are active during mental imagery and motor execution of swallowing. This task consisted of 4 motor execution trials and 4 mental imagery trials. The data of the localizer session was analyzed offline using Brain Voyager QX v.2.3.1 and used to identify the region of interest (ROI) for the NF task, which was performed during the second session. The ROI for the NF task was extracted individually for each participant. Clusters of activity in the left lateral precentral gyrus were observed in all participants. During the second session, participants received real-time feedback of activation changes in the left lateral precentral gyrus using Turbo-BrainVoyager. Each run included 10 resting trials and 10 NF trials. All trials had a duration of 30 s. Functional images were acquired using a T2* weighted gradient-echo pulse imaging sequence (TR = 2400 ms; TE = 30 ms; flip angle = 90◦; matrix = 68 x 68; slice thickness = 3.5 mm; voxel dimensions = 3.5 x 3.5 x 3.5 mm) providing whole brain coverage in 36 slices. Anatomical images were recorded using a T1-weighted MPRAGE sequence (TR = 2530 ms; TE = 2.26 ms; flip-angle = 9°; slice thickness = 1 mm; 256 x 256 acquisition matrix; voxel dimensions = 1 x 1 x 1 mm; TI = 900 ms). SPM 8 was used to preprocess and analyze functional whole brain data with the purpose of analyzing group effects. The derived spatial transformation was then applied to the realigned T2∗ volumes, which were spatially smoothed with a Gaussian kernel of 8-mm FWHM.   Results: Localizer: To examine brain activation patterns during ME and MI of swallowing, we contrasted both conditions with the resting condition. During ME of swallowing, a large network of brain areas was active including the bilateral cerebellum, bilateral pre- and postcentral gyrus, basal ganglia, the insula, motor areas and the SMA. During MI of swallowing when no real-time feedback was provided (MI_offline), comparable brain areas were active than during ME of swallowing (Table 1). Neurofeedback training: Only the right precuneus showed a stronger activation during the second compared to the first run (voxels: 21, peak: x: 2, y: -64, z: 54, T-value: 5.94). Activity levels observed during the mental imagery offline task and the first feedback run in the feedback ROI correlated significant positively (r = 0.61, p < 0.05). Hence, the higher the activation during the offline task, the higher the ability to up-regulate activation during feedback.   Conclusions: During neurofeedback training, participants were able to increase the activity in the left lateral precentral gyrus and in other brain regions, which are generally active during swallowing, compared to the motor imagery offline task. Our results indicate that motor imagery of swallowing is an adequate mental strategy to activate the swallowing network of the whole brain, which might be useful for future treatments of swallowing disorders.

この発表は， 画像による動きのイメージと実際の実行時の脳の運動領域の活動の比較であった．これは，fNIRSで手や足の動きで研究されていた内容をfMRIを用いて脳全体で解明しようとするものであった．この研究では，実行時のほうがイメージしているときよりも多くの脳領域が活動的であったが，left-IFGはイメージ時のほうが強い活性が見られたという結果であった．私は，fMRIの研究をfNIRSに適用することは多いように思っていたが，逆のパターンであったため興味を持った．

発表タイトル       ：Abstract and concrete conceptual representation in the inferior parietal lobe: fNIRS 著者                  ： Maria Montefinese1,2, Paola Pinti3,4, Ettore Ambrosini1,5,6, Ilias Tachtsidis3, David Vinson2 セッション名       ： Poster Session Abstruct            ： Introduction: Similarity measures can inform us about the nature of semantic representation: the knowledge we have of the world. Numerous theories of semantic representation exist, some based upon our sensorimotor experience as inferred from verbal features (e.g. featural similarity) or property ratings (e.g. affective content), others based on regularities in spoken and written language (e.g. lexical co-occurrence) [1]. Similarity measures may also be estimated from lexical measures (e.g. orthographic similarity) [2] and behavioural studies (e.g. word association) [3].  A meta-analysis of 120 fMRI studies showed a left lateralized brain network in semantic processing [4], including the angular and supramarginal gyri. In particular, the left inferior parietal lobe (IPL) is one of the most informative regions for abstract and concrete word classification [5].  However, no neuroimaging study has investigated which similarity measure best predicts patterns of brain activity in IPL, and whether this differs for abstract and concrete words. Here, we test this using fNIRS [6], which ensures a higher degree of ecological validity than fMRI. We expect the IPL to code different kinds of similarity measure depending on the word concreteness.   Methods: 13 native English speakers (9 M, mean age: 26.7 years) performed a semantic decision task (is a word abstract or concrete?) on 160 visually presented words (80 abstract, 80 concrete) (Figure 1A). Stimuli were repeated over 6 runs. Hemodynamic changes in IPL were monitored bilaterally using a 20-channel fNIRS system (LIGHTNIRS, sampling rate = 13.33 Hz; Figure 1B). Channels’ locations were digitized and co-registered onto a standard MNI brain template. fNIRS data were corrected for motion artifacts and band-pass filtered (.01-.6 Hz) to remove physiological noise. Oxy- and deoxy-hemoglobin (HbO2 and HbR) signals were then down-sampled to 3 Hz and analyzed with a General Linear Model (GLM) approach [7] to derive the beta-estimates for each word. This was carried out on the channels covering the IPL (green-filled circles in Figure 1B).  The analysis involved: (i) a Representational Similarity (RS) Analysis [8] based on Spearman partial correlations, performed between the similarity matrices based on both neural activity patterns (Brain) and similarity measures across abstract and concrete concepts (Models). HbO2 and HbR were analyzed separately. We derived 5 theoretical models based on featural similarity, word association, lexical co-occurrence, affective content, and orthographic similarity (control model); (ii) a leave-one-out item-level multivariate classification analysis, carried out using the procedure in [9], to decode the concreteness category of single words (trial-level decoding). Subject-wise RSs and mean trial-level decoding accuracies were compared between hemispheres and abstract and concrete categories with GLM analysis.   Results: For HbO2, we found different Brain-Model RSs for co-occurrence between abstract and concrete words, regardless of the hemisphere. This was due to a significant RS for concrete words only. Moreover, we found different Brain-Model RSs for affective content between hemispheres and word concreteness, with a significant Brain-Model RS for abstract words in the left hemisphere only. No significant Brain-Model RS were found for HbR, probably due to HbR smaller changes than HbO2 and consequent lack of statistical power [6].  For HbO2, we also found lower accuracies for word association-based decoding for concrete words as compared to abstract ones, regardless of the hemisphere.   Conclusions: These results indicate that IPL represents semantic-affective information depending on word concreteness, implying that semantic representations of abstract and concrete words are governed by different organizational principles. Moreover, they suggest the feasibility of multivariate methods as a tool for fNIRS decoding of pattern-based neural correlates of cognition.

この発表は， LIGHTNIRSを用いて研究されていました．本研究は，提示された単語に対して意味決定タスクを実行することにより，神経活動パターンと単語の類似度を測定するものでした．また，多変量パターン解析（Multivariate pattern analysis: MVPA）[2]により，グループ解析がされていました．この手法の実現可能性を示唆しており，今後の研究に使用することは可能か調査したいと思いました．
 

発表タイトル       ： Monitoring Brain Activity during Rhythmic Music Therapy: an fNIRS Investigation 著者                  ：Sabrina Brigadoi1 , Federico Curzel1 , Simone Cutini1 セッション名       ： Poster Session: NIRS Abstruct ： Introduction: Music therapy is a method used in movement disorders, such as Parkinson disease (Thaut et al., 1996), relying on the influence of rhythmic stimulation on movement coordination (Chen et al., 2006; Palomar-García et al., 2016). Functional near-infrared spectroscopy (fNIRS) is a non-invasive and portable optical imaging technique that can be used to monitor brain activity in tasks involving participants’ movements (Cutini et al., 2014). In this study, we used fNIRS to monitor brain activity of healthy participants whilst performing a rhythmic music therapy session playing the drums. The aim was to detect modulations of hemodynamic activity in areas involved in music perception and movement control depending on the type of exercise performed to elucidate the mechanisms of movement facilitation.   Methods: Twenty-six healthy volunteers (mean age 21.27±2.74 years) performed a paradigm composed of 8 parts. The first and the eighth part were a passive listening of a track with a harmonic and rhythmic line, the other parts were rhythmic-motor tasks. Ten blocks of 18 s were administered for each task type, followed by a rest period (13-18 s). Participants were asked to play two MIDI pads placed in front of them (Fig. 1a). In the second task (ex2), participants were asked to produce a constant rhythm of 1 Hz with no external support. In the thir For each participants and task condition, the average time interval between consecutive beats was computed. The difference between this interval and 55 bpm was computed as behavioral metric. Hemodynamic data were acquired with the ISS Imagent system (8 detectors-32 sources). A symmetric probe was created covering motor-temporal areas, with a total of 44 standard channels (3 cm) and 2 short-separation (SS) channels (Fig. 1b).Homer2 (Huppert et al., 2009) was used to preprocess the data using spline interpolation and wavelet methods for motion artifact correction and a band-pass filter (0.01-0.5 Hz). SS channels were used to remove physiological noise. The average value of the HRF in the interval 5-15 s from stimulus onset was computed and submitted to statistical analyses. Both behavioral and hemodynamic data were analyzed with repeated measures ANOVAs, followed by paired t-tests.   Results: Participants improved their rhythmic ability during and after the training (F(6,192)=37.476; p<.001), producing in ex7 a closer rhythm to the expected one than in the first two tasks (ex2 vs. ex7: t(32)=4.49, p < .001; ex3 vs. ex7: t(32)=3.16, p = .003; Fig. 2a).  Hemodynamic data showed a different involvement of the motor and temporal areas during the different tasks, with less channels being activated during ex5, ex6 and ex7 compared to the initial tasks. Type of task had a significant impact on modulation of hemodynamic activity depending on the channel (F(126,3150)=1.557, p < .001). Seven channels broadly located in the supplementary motor area (SMA, Fig. 2b,c,d) resulted to be modulated depending on the task.   Conclusions: Results demonstrate that the proposed paradigm modulates brain activity mainly in the SMA, suggesting that this could be a suitable paradigm to test the effects of music therapy on patients with pathologies affecting movements. SMA was found to have a fundamental predictive role in motor synchronization (Chen et al., 2006). Furthermore, our results suggest the importance and power of rhythm in movement facilitation, how melody can further strength this effect and our ability to internalize the rhythm once learnt. Next step will be the use of this paradigm on Parkinson’s patients to understand how music therapy can modulate brain activity in their “damaged” brain, aiming to achieve an individual treatment.

この発表は， fNIRSを用いて運動促進のメカニズムを明らかにするために行われ，運動の種類に応じて音楽の知覚と運動制御に関する領域の血行動態活動の変調を検出したものでした．Homer2を用いて解析されており，体動除去手法にはスプライン補間とwaveletの2種類が用いられていました．また，賦活解析には刺激後5-15[s]の間のHRFが用いられていました．本研究では運動にはSMAが影響を与えることが示されていた．運動に関する研究であり，同じ解析ソフトが用いられていたため，自身の研究の参考にしていきたいと思いました．
 
参考文献
[1]　OHBM2019 Annual meeting,
https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageID=3882&activateFull=true
 
[2]　 Emberson, Lauren L., et al. “Decoding the infant mind: Multivariate pattern analysis (MVPA) using fNIRS.” PloS one12.4 (2017): e0172500.
学会参加報告書

報告者氏名	大塚友樹
発表論文タイトル	安静時および瞑想時における機能的接続性の個人間および個人内変動の検討
発表論文英タイトル	Intra- and inter-individual variation in the resting- and meditative-state functional connectivity
著者	大塚友樹, 日和悟, 廣安知之
主催	Organization for human brain mapping
講演会名	OHBM2019
会場	Auditorium Parco Della Musica
開催日程	2019/06/09-2019/06/13

 
 

1. 講演会の詳細

2019/06/09から2019/06/13にかけて，ローマにて開催されました2019 OHBM annual meetingに参加しました．この2019 OHBM annual meetingは，Organization for human brain mappingによって主催された国際会議で，ヒトの脳組織および脳機能のマッピングに関する研究に携わる様々な背景を持つ研究者を集め，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^（１．本研究室からは他に，廣安先生，日和先生，M2の学生として古家，奥村(駿)，吉田，山本，杉野，M1の学生として風呂谷，丹が参加しました．
 

2. 研究発表
   • 発表概要

私は11日のポスターセッションおよびポスターレセプ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，計3時間参加者の方と議論を行いまいした．
今回の発表は，「Intra- and inter-individual variation in the resting- and meditative-state functional connectivity」について発表いたしました．以下に抄録を記載致します．

Introduction Conventional studies on the neural basis of mindfulness meditation did not consider the intra-individual variation in the brain state, as they assumed it to be small compared with the inter-individual variation. However, the brain functional connectivity (FC) networks present variability within an individual as well. As shown in Fig. 1, result reproducibility would be poor in a brain region showing larger intra- than inter-individual variation, as the group analyses would be strongly influenced by intra-individual variation. Therefore, inter- and intra-individual variation during meditation were investigated in the present study through repeated measures of the same participant on different days to determine the reliability of the estimates on the brain functional network involved in meditation.   Methods A total of 21 healthy novice meditators (22.9 ± 2.5 years; 6 females, 15 males) performed breath-counting meditation for 5 minutes, which consisted of the mental counting of breaths, following a 5-minute resting state in an fMRI scanner. Specifically, the same task was performed 10 times on different days in 3 subjects (i.e., subjects A, B and C). Furthermore, the FC matrix among the brain regions was calculated for the 116 regions defined by the automated anatomical labeling (AAL). In addition, the Pearson’s correlation coefficient between the FC matrices of participants in the group of novices and those of participants A, B, C during both the resting and meditative state was calculated and defined as the similarity matrix. The similarity matrix was converted into the dissimilarity matrix. It was used for the distance measure of classical multidimensional scaling (MDS) to visualize the FC matrices during the resting and meditative state on a two-dimensional space (M.L. Davison 1983). Finally, the variation of the functional connectivity matrix between the experimental conditions and within individuals were compared.   Results As illustrated in Fig. 2, the brain functional connectivity matrix during both the resting and meditative states were visualized for each subject implanted in the 2D space by MDS and the distance between the 21 participants of the novice group and the 3 subjects (i.e., A, B, C) was measured. While the solid ellipse represents the meditative state, the dotted ellipse shows the 95% confidence interval in the resting state bivariate normal distribution, whereas the comparison between the intra-individual variation of the subjects A, B and C and the group variation is displayed in Fig. 2 (b-d). These results indicate a smaller intra-individual variation of subjects A, B, C within the resting and meditative states during the day than the inter-individual one of the group. Our finding suggests that participants’ inter-individual variation has a greater influence on the brain states than the daily fluctuation. Furthermore, as reported in Fig. 2 (e), the distance between subjects at rest and during meditation was significantly larger (p <0.05) than that between the rest and meditation states in the same subject. This result revealed that the inter-individual variation in both the resting and meditative states is greater than the change in FC networks associated with meditation. Conclusion In the current study, the intra- and inter-individual variation in the brain functional network involved in meditation were investigated using MDS. Our results indicated that the inter-individual variation in the resting and meditative states is significantly larger than the fluctuation from resting to meditation in the same individual. This suggests that the variation among individuals is larger than that associated with the quality of meditation in a given individual.

 

• 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，被験者は瞑想実践者か初心者かというものでした．この質問に対しては初心者ですと答えました．
 
・質問内容2
質問は，どうやって解析しているのかというものでした．この質問に対しては，各被験者のFC matrixをMDSを用いて二次元に落とし込みそれぞれのFC matrixの距離からばらつきを検討していますと答えました．
 
・質問内容3
質問は，どうやって低次元にマッピングしているのかでした．この質問に対しては，距離行列を基にしたMDSを使ってマッピングしたと答えました．
 
・質問内容4
質問は，restとmeditationは違うのかというものでした．この質問に対しては，距離行列行列から類似度が高いと答えました．
 
・質問内容5
質問は，距離行列を求める手法は何か文献を参考にしたのか，それとも，オリジナルなのかというものでした．この質問に対しては，Functional Brain Networks Are Dominated by Stable Group and Individual Factors, Not Cognitive or Daily Variationの論文を参考にしたと答えました．
 
・質問内容6
質問は，瞑想は何を行なっているのかというものでした．この質問に対しては，数息観を行なっていると答えました．
 
・質問内容7
質問は， MDSのプロットをどうやって定量化したのかというものでした．この質問に対しては，低次元に落とし込む前の距離行列を基に定量化をしたと答えました．
 
・質問内容8
質問は，質問紙やbehaviorは取っているのかというものでした．この質問に対しては，質問紙は取っているがbehaviorは取っていないと答えました．
 
・質問内容9
質問は，個人の特徴とは何かというものでした．この質問に対しては，性別や年齢などと答えました．

• 感想

・去年に引き続き，2度目のOHBMでの発表でした．去年とは異なり，1日のみの発表でしたが，去年よりも多くの方が聴きに来てくださり，たくさんの方と議論をすることができました．ばらつきがテーマの研究が多くあり，自身の研究テーマが世界から注目を集めていることを実感しました．今回は，発表までの準備が不十分であり反省する必要があると思いました．計画を立てることの必要性を感じました．今後は，今まで以上にwbsを活用し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2件の発表を聴講しました．
 

発表タイトル       ：　Individual Variability of Functional Connectivity in Resting-State and Naturalistic fMRI Paradigms 著者                  ： Mark O’Reilly セッション名       ： Poster session Abstruct            ： Introduction：　Resting state fMRI studies are often criticized due to their lack of control over the cognitive states of individuals during observation, which may lead to increased intersubject variability (ISV) in estimates of functional connectivity. Mueller and colleagues (2013) estimated the ISV of the functional connectivity architecture in the human brain at rest, finding higher variability in heteromodal association cortex and lower variability in unimodal areas such as primary sensory or motor cortex. Engaging movies have been shown to increase intersubject correlations (ISC) of neural activity in sensory and other regions (Hasson 2010), suggesting an alignment of cognitive states across individuals based on the events of the movie, potentially reducing intersubject variability in connectivity estimates. The objective of this study is to investigate the differences in ISV of functional connectivity between rest and movie conditions. Methods: Minimally preprocessed 7T data of 25 subjects (25 F, Age 31-35) from the Human Connectome Project (Van Essen 2013) were analyzed, with each subject providing four 15-minute resting state sessions and four 15-minute movie stimulation sessions. The Glasser parcellation (Glasser 2016) was used to identify 180 regions of interest (ROI) per hemisphere in each subject for further analysis. A functional correlation matrix was computed for each subject and each session (100 total), where each row or column represents the connectivity fingerprint, a 1 x 360 vector of correlations, for a particular ROI. For each ROI, the variability across the 25 subjects was quantified by averaging the correlation values between each pairing of subject fingerprints and inverting the averaged value to attain a measure of dissimilarity. This was performed for each session and averaged across sessions to get a single measure of dissimilarity. To remove effects of measurement noise, intrasubject variability was calculated using a similar method but using dissimilarity across sessions, instead of subjects. The intrasubject variability was regressed from the intersubject variability to obtain a final variability value. This process was performed in both conditions to obtain two variability maps. The movie variability was subtracted from the rest variability to highlight differences between the conditions. Finally, we speculated that ISC in the movie might explain reduced ISV, so ISC values of the movie BOLD timeseries were calculated by region to determine the overlap between ISC and differences in variability of functional connectivity between conditions. Results: To test if ISV differences between conditions were associated with ISC values in the movie condition, we computed the correlation between the two variables and found them to be moderately correlated across the brain (r=0.554, p<0.0001). The variability difference map is presented in Fig. 1. Regions with notable differences between the conditions include prefrontal and superior temporal cortex. Differences observed in the superior temporal cortex may suggest that the movie induces more stable connectivity in structures involved in speech processing and auditory sensation across participants than the rest condition. The differences observed in the prefrontal cortex may suggest a similar conclusion for structures involved in high level cognitive processes. When we compare the difference map to the ISC map shown in Fig.2, we observe that high levels of ISC are present in the superior temporal, but not in prefrontal cortex. We validate MN-SRM (a hyperalignment method) on real data: our ECM algorithm estimates far fewer parameters than SRM by integrating the projection matrices instead of the shared time series. As a result, we show modest improvements in out-of-sample reconstruction relative to SRM on the sherlock and raider datasets (Chen et al., 2016; Haxby et al., 2011). Conclusions: We analyzed the spatial distribution of differences in intersubject variability between the rest and movie conditions. We also found that variability differences exist in superior temporal and prefrontal cortex. We also found an association between these differences and ISC values in the movie condition. This suggest that the high ISC in the movie may result in more stable estimates of functional connectivity across subjects.

この発表は，安静時と映画を見ている時の個人間と個人内のばらつきを検討していました．解析手法が自分のものと似ており，自分の研究では，FC matrix間の相関を求めているが，この研究では，領域ごとに被験者間と被験者内のセッション間の相関を求めていました．自分の研究でも試してみる価値があると考えています．
 

発表タイトル       ：Hemispheric Difference in Group, Task and Individual-dependent Variation of Functional Networks 著者                  ： Chenxi Zhao, Yaya Jiang, Xinhu Jin, Gaolang Gong セッション名       ： Poster session Abstruct            ： Introduction： As a topic of general interest, individual difference of human brain has been intensively studied[1,2]. By comparing human with monkeys, a previous study has shown human-specific left-lateralized anatomical variations[3], suggesting a hemisphere-dependent variation in particular brain phenotypes. Recently, the contribution of the group, task, and individual factors to the variation in whole-brain functional networks has been detangled[2]. Following this, the present study aims to examine the magnitudes of group, task and individual-dependent variations in the two hemispheric functional networks, and further evaluate the hemispheric differences. In addition, the heritability of the hemispheric differences in each type of variation was assessed. Methods: In total, 933 human connectome project (HCP) subjects (508 females,age: 22-37,212 monozygotic twins) with resting-state and task fMRI scans(emotion, language, motor, working memory) were included. All images were preprocessed by the HCP pipeline and then feed into the GRETNA toolbox[4] to do linear detrending(only resting-state fMRI), nuisance signals regression and temporally filtering (resting-state:0.01–0.1 Hz, tasks:>0.01 Hz). The AICHA atlas[5] was used to define the nodes of network(186 in each hemisphere). Functional connectivity between each within-hemispheric node pair was defined by the Pearson correlation of mean time series (z transformed). For each hemisphere, a network similarity matrix was calculated by correlating among the linearized upper triangles of hemispheric network matrices(Fig1A). As did by Gratton et al.[2], the group, task, and individual-dependent variations were calculated per subject as following: 1) the average similarity from different individuals and tasks(group, baseline), 2) the added similarity from the same task but different individuals relative to group(task), and 3) the added similarity from the same subject but different tasks relative to group(individual). The task and individual-dependent variations were compared with the group-dependent variation using paired t-tests. To test the hemispheric differences, two hemispheric variations attributable to each factor were compared using a paired t-test. For each significantly lateralized effect, the asymmetry index (AI=(L-R)/(L+R)) was calculated and its heritability (h2) was estimated using the SOLAR software[6]. Multiple comparisons were corrected by the Bonferroni method (p<0.05). Results: As shown in Fig1, both left and right hemispheric (LH and RH) functional networks showed substantial similarity across group (LH/RH:0.46±0.03/0.44±0.03) and added similarity of networks from the same individual (LH/RH:0.18±0.08/0.17±0.07), whereas subtle but significant added similarity due to task (LH/RH:0.09±0.04/0.1±0.04). Paired t-tests showed significant left-lateralized contributions of the group (t=26.8,p=0) and individual (t=4.9,p=10-6) factors to network variation, but a right-lateralized contribution of the task factor (t=20.1,p=0) (Fig2). As listed in Table1, the h2 was significant for the AI of group (h2=0.18,p=0.002) and individual-dependent variation (h2=0.24,p=7×10-5) but non-significant for the task (h2=0.18,p=0.027, not surviving the multiple-comparison correction). Conclusions: Our results demonstrated a strong hemispheric functional network stability (group-shared organization and individual features) and moderate state-dependence. Intriguingly, the variation attributable to either the group, task, or individual factors markedly differed between the two hemispheres: shared group-level factor and individual-specific features had stronger influences on the LH network organization, while state-changes had a greater impact on the RH network. Furthermore, our heritability results indicated a significant genetic role in hemispheric differences in group and individual-dependent variations, though tenuously. These findings together provide novel insight into the hemispheric functional network organization and its lateralization.

この発表では，HCPのデータセットを使用しており被験者数が933人でタスクと個人特性における脳機能ネットワークのばらつきについて検討していました．Taskは，emotion, language, motor, working memoryを使用していました．結果は，脳機能ネットワークはタスクよりも個人特性の影響を受けるとされていました．933人という多くの被験者でそのことが示されていました．
 

発表タイトル       ： Individual-Specific Areal-Level Parcellations Improve Functional Connectivity Prediction of Behavior 著者                  ： Ru Kong, Qing Yang, Evan Gordon, Xinian Zuo, Avram Holmes, Simon B. Eickhoff, B. T. Thomas Yeo セッション名       ： Poster session Abstruct             ： Introduction：　A The human cerebral cortex comprises hundreds of functionally distinct areas, which are in turn organized into at least ten to twenty large-scale networks. Most resting-state fMRI (rs-fMRI) parcellations have relied on group-averaged data[1–3], which might obscure individual-specific topographic features[4, 5]. Here, we propose an approach to generate individual-specific areal-level parcellations and show that the resulting parcellations can improve individual predictions of behavioral phenotypes based on functional connectivity (FC). Methods: We have previously proposed and validated a multi-session hierarchical Bayesian model (MS-HBM) for estimating individual-specific cortical networks[6]. The multiple layers of the MS-HBM explicitly differentiate intra-subject (within-subject) from inter-subject (between-subject) network variability. In contrast to MS-HBM, previous network mappings ignore intra-subject variability, so might confuse intra-subject variability for inter-subject differences.   We have previously utilized MS-HBM for network parcellations[6].To estimate areal-level parcellations, the MS-HBM could be re-trained by initializing with a group-level areal-level parcellation (e.g., Schaefer2018[7]). Furthermore, we constrained the individual-specific parcels to be within 30mm of the group-level parcels, since previous studies suggest that individual variation in cortical areal location can go up to 30mm[8].   We compared MS-HBM with a well-known individual-specific parcellation approach (Gordon2017[5]). We considered rs-fMRI from 10 subjects (10 sessions each) in the MSC dataset[5]. Each subject was parcellated using all rs-fMRI sessions. Task inhomogeneity[5, 7] (standard deviation of task activation within each parcel) was then evaluated using task-fMRI data from the same subjects. A lower task inhomogeneity indicates better parcellation quality. For fair comparison, the number of MS-HBM parcels were constrained to be the same as Gordon2017.   Second, we considered ICA-FIX denoised rs-fMRI data from the HCP S1200 release[9]. We selected 58 behavioral measures across cognition, personality and emotion[6]. Individual-specific MS-HBM parcellations were estimated for subjects with four runs and no missing behavior (N = 752). For each subject, we obtained 400×400 FC matrices using the 400-area group-level parcellation (Schaefer2018) or the 400-area individual-specific MS-HBM parcellations. The FC matrices were then used for predicting the 58 behavioral measures using kernel ridge regression[10]. We performed 20-fold cross-validation: kernel ridge regression was trained on 19 folds and used to predict behavior in the test fold. The regularization parameter was determined using inner-loop cross-validation. Furthermore, the 20-fold cross-validation was repeated 100 times[6]. Results: Individual-specific MS-HBM parcellations achieved better task inhomogeneity than Gordon2017, suggesting better generalization to task data (Fig. 1A). Fig. 1B shows the parcellations of two representative MSC subjects estimated from 5 rs-fMRI sessions. We observed significant topological differences between the two subjects, which were highly replicable across sessions.   Compared with Schaefer2018, the FC of MS-HBM parcels achieved a higher average prediction accuracy with a relative improvement of 9.44%. We note that we could not compare with Gordon2017 because the Gordon2017 approach estimated different number of parcels in each participant, so the resulting FC matrices were not comparable across participants. Conclusions: Compared with other approaches, MS-HBM individual-specific cortical parcellations generalized better to new rs-fMRI (not shown due to space constraints) and task-fMRI data from the same subjects. MS-HBM parcellations were highly reproducible within individuals, while capturing unique individual features. Individual-specific parcellations yield better FC-based behavioral prediction compared with group-level parcellations.

この発表は，MSCデータセットを使って自身が提案したパーセレーションの手法を適応していました．個人内において，日による変動が小さいことがパーセレーションされた脳においても示されていた．

発表タイトル       ：Individual parcellation of structural connectivity with machine learning 著者                  ： Hantian Zhang, Lingzhong Fan, Luqi Cheng, Tianzi Jiang セッション名       ： Poster session Abstruct            ： Introduction：　Individual parcellation is important for clinic and necessary for understanding brain. However, the performance of single-subject parcellation is limited (Wang D et al.2015;Chong M et al.2017). Structural connectivity is widely used to parcellate cortex according to different connectivity profiles. Typical pipelinex(Fan L et al.2016) employed volume-based diffusion tractography and used spectral clustering to parcellate each subject separately. The limited data for single-subject may obtain instable result(Chong M et al.2017). In order to improve tracking performance and acquire a model collecting information of group-average atlas and existing dataset. We apply surface-based tractography and performed a supervised machine learning method(Glasser, M et al.2016; Ganepola, T et al.2018) based on structural connectivity.   Methods: In the current study, we parcellate right cortex into 105 regions using surface-based probabilistic tracking and supervised learning. We choose Brainnetome Atlas(Fan L et al.2016) as our template.180 subjects for training were obtained from Human Connectome Project(HCP) database. Another 30 subjects were used from HCP test-retest data to evaluate the results. There is no overlap between the two datasets and 40 HCP subjects selected by Brainnetome atlas doesn’t overlap with these two datasets.MRI data were preprocessed by Freesurfer Pipeline HCP script to generate surface mesh. Each cortex region’s probability map(PM) was projected into corresponding position on group-average surface mesh. Then all PMs were merged on surface. Merged PM was employed as a measurement of inter-subject variability. Probabilistic tracking was performed and white matter surface was selected as seed. Connectivity profile was represented the distribution of fibers connecting brain regions labelled with Brainnetome Maximum Probability Map(MPM).Firstly, 180 HCP subjects were divided into training (N=150) and validation(N=30).Labels from Brainnetome MPM were (Fig. 1A)mapped into individual surface mesh as training labels. Secondly, the vertices with high inter-subject measured by merged PM(Fig. 1B) were filtered with a threshold and remaining vertices were used for training. The step aims to avoid vertices-level misalignment of label across template and single-subject.Thirdly, Areal classifiers(Fig.1C) of all regions in cortex were trained as region(blue) against neighbors(yellow) using multi-label Random Forest(max_depth=10, n_estimators=150). Finally, when a new individual comes, subject-specific results were predicted with trained areal classifiers(Fig.1D). Dice coefficient was computed for test-retest dataset to quantitatively analysis reproducibility and variability. Results: 105 areal classifiers were trained for right cortex. Mean accuracy was 92% obtained in validation set and the lowest accuracy still got 88%.Inter-subject similarity was measured as mean pairwise dice from HCP-test dataset and intra-subject similarity was calculated in test-retest dataset. Inter-subject dice(0.51) is lower than intra-subject dice(0.76) revealed in Fig.2B. 4 subject-specific parcellation results (without post processing) from HCP test-retest dataset are displayed in Fig.2A.Two arrows pointing two representative parcels(A44rd,A40d) which are variable across subjects while relatively robust within subjects. Conclusions: The proposed method obtained discriminant classifiers for each region. The parcellation method achieved intuitive result about inter-subject and intra-subject variability. We need more caution to choose the threshold to filter vertices while remaining enough correct vertices for learning. In the future, we will test distinct quality-level dataset and train different areal classifiers with specific parameters and strategy.

この発表は，機械学習を用いて脳の構造的接続性における個人脳パーセレーションを行なっていました．被験者間における類似度は被験者内の類似度よりも低いことから脳構造は被験者によって異なることが明らかになっていました．
 
学会参加報告書

報告者氏名	山本渉子
発表論文タイトル
発表論文英タイトル	An fMRI study on the attentional state induced by breath-counting meditation
著者	山本渉子, 日和悟，廣安知之,
主催	Organization for Human Brain Mapping
講演会名	OHBM2019
会場	Auditorium Parco Della Musica
開催日程	2019/06/09-2019/06/13

 
 

1. 講演会の詳細

2019/06/09から06/13にかけて，ローマのAuditorium Parco Della Musicaにて開催されましたOHBM2019に参加致しました．OHBM2019は，Organization for Human Brain Mappingによって主催された学会で，人間の脳のマッピングにおける装置を横断する最新の国際的な研究について学ぶための場所です．また，この分野の専門家と議論し，世界中とつながることができる機会です．教育セッションでは，様々な経歴を持つ若手および上級の科学者が，機械学習技術，高解像度画像処理，そして最近ではオープンサイエンス手法を含む，この分野の最新，かつ画期的な開発について教えています．
私は全ての日程に参加致しました．本研究室からは他に廣安先生，日和先生，M2古家くん，大塚くん，奥村(駿)くん，杉野さん，吉田さん，M1風呂谷くん，丹さんが参加しました．
 

2. 研究発表
   • 発表概要

私は6/11の12時45分から14時45分に行われた「Poster Session」，17時から18時に行われた「Poster Reception」にて発表致しました．発表の形式はポスター発表で，計2時間でした．
今回は，An fMRI study on the attentional state induced by breath-counting meditationというタイトルで発表致しました．以下に抄録を記載致します．
 

[Introduction] Mindfulness refers to paying attention to the present moment, nonjudgmentally, and is expected to promote individuals’ mental well-being. Considering that attention is one of the main components of meditation, practitioners try to sustain their attention on their physical sensations during meditation to improve their interoceptive attention. In this study, we aim to investigate whether meditation induces individuals’ interoceptive attention through the fMRI scanning of brain activity during breath-counting meditation. Furthermore, the exteroceptive state of practitioners observed while responding to and counting the external cues was assessed to emphasize the characteristics of the interoceptive state. The brain activity seen in such states was investigated using activation analysis. [Methods] A total of 12 male meditation novices (aged 23.1 ± 1.2 years), who never experienced meditation practices, participated in the experiment. As reported in Fig. 1, they were asked to perform two tasks, namely a breath-counting task (BCT) and an auditory counting task (ACT). The BCT was first proposed by Levinson et al. and was shown to be effective as a behavioral measure of mindfulness meditation (Levinson et al. 2014). In the BCT, participants repeatedly counted their breaths mentally from 1 to 9 and pressed two buttons at breaths 1–8 and at the ninth breath, respectively. In contrast, they were requested to quickly press the button responding to the auditory cues in the ACT and to count them similarly to the BCT. Furthermore, they restarted counting from 1 in both tasks when they got distracted and pressed a third button. An fMRI scanner was used to measure the brain activity during such tasks. The whole brain was first divided into 116 regions based on the automated anatomical labeling atlas. Successively, an activation analysis was performed on the measured data after the preprocessing, which included realignment, slice timing, coregistration, normalization, and smoothing by SPM12. Finally, a group analysis was performed on the data of all subjects. [Results] Frontal_Mid_R (MFG.R) was significantly activated in the ACT (uncorrected, p<0.001). Since the MFG is included in the central executive network and is known to be involved in external-oriented tasks (Manoliu et al. 2013), attention was suggested to be directed to the outside of the body in the ACT. Moreover, as described in Table 1, brain activation was increased in the Insula_L/R (INS.L/R), Temporal_Sup_L/R (STG.L/R) during the ACT compared to the BCT (uncorrected, p<0.001). While the INS is part of the salience network (SN) and is related to the sensing of external stimuli, the STG is associated with the recognition of auditory information. Therefore, contrary to the BCT, attention was indicated to be directed to the outside as an auditory stimulus in the ACT. Based on these results, the characteristics of the brain activity of the BCT were investigated by comparing them with those of the ACT. In contrast, brain activation was increased in the Cingulum_Ant_L (ACG.L) in the BCT compared to the ACT (uncorrected, p<0.001), as illustrated in Table 2. Since the ACG is included in the SN and is related to both self-recognition and attentional control, attention was suggested to be directed to the bodily sensations in the BCT as opposed to the ACT. Overall, we assumed that meditation induced participants’ interoceptive attention. [Conclusions] In this study, the brain activity of meditation novices during a task that focused on the attention to the self was investigated by comparing it with that observed during a task in which attention was directed to the outside the body. As a result, brain activation was increased in the ACG (i.e., the region related to self-recognition, attentional control) during meditation (i.e., attention to the physical sensations) compared to the task in which the attention was directed to external stimuli. These results suggest that meditation induced participants’ interoceptive attention. 【References】 [1] D.B. Levinson, E.L. Stoll, S.D. Kindy, H.L. Merry and R.J. Davidson (2014), ‘A mind you can count on: validating breath counting as a behavioral measure of mindfulness’, Frontiers in psychology, vol. 5, p.1202. [2] Manoliu, Andrei, et al (2013), ‘Aberrant dependence of default mode/central executive network interactions on anterior insular salience network activity in schizophrenia’, Schizophrenia bulletin, vol. 40, no.2, pp. 428-437.

 

• 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，feature valuesとは何かというものでした．この質問に対して，今回はグラフ理論の指標であるdegree centlarityとbetweenness centralityを使用したと回答しました．また，この特徴量はなにかという質問もいただいたので，degree centlarityは，他の脳領域との接続数であり，betweenness centralityは，他の2つの領域を結ぶ最短経路にその脳領域が含まれる割合であると説明致しました．
 
・質問内容2
質問者の氏名を控え損ねてしまいました．こちらは，BCT，ACTそれぞれの実験方法は自分のオリジナルの方法なのかという質問でした．この質問に対して，BCTは既存の手法を用い，ACTは自分で考えたものであると回答しました．
 
・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，Counting accuracyとはなにかというものでした．この質問に対して，実験中どのようにボタンプレスを行っているかを説明し，正しいボタンプレスのセット数を総セット数で割ることで算出していると回答しました．
 
・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，MWとは何を表しているのかというものでした．この質問に対して，マインドワンダリングの略称であると回答しました．
 
・質問内容5
質問者の氏名を控え損ねてしまいました．この質問は，結果の脳の図に対して領域名を複数記載していたため，１つのnodeの領域名は何かというものでした．この質問に対して，Temporal Inf R（ITG.R）であると回答しました．領域名の略称については他の方からもご質問いただいたので，正式名称をお伝えしました．
 
・質問内容6
質問者の氏名を控え損ねてしまいました．こちらの質問は，このタスクは人によって数える対象の総数は異なるのかというものでした．この質問に対する回答ですが，両タスクで個人によって回数が異なり，BCTは人によって呼吸のスピードが違うため数は異なると回答しました．また，呼吸のスピードは測定しているのかという質問もいただいたので，行動データについてはボタンプレスのみ記録しているとお答えしました．
 
・質問内容7
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，ボタンプレスの際，緑のボタンはACTでも押すのかというものでした．この質問に対する回答ですが，ACTでもカウントを間違えたり，気が逸れたら押すよう指示していると回答しました．
 
・質問内容8
質問者の氏名を控え損ねてしまいました．こちらは，BCTでは瞑想を行っているのにマインドワンダリングになっていいのかという質問でした．この質問に対して，先行研究において集中瞑想時には認知モデルのサイクルが存在すると言われており，マインドワンダリングになっている状態と集中できている状態が両方存在することを説明しました．また，今回の実験では，被験者が全員瞑想初心者であったため，今後は瞑想実践者も測定し，初心者と比較していきたいとお伝えしました．
 
 

• 感想

OHBMには，私たちの研究室から毎年参加していますが，私は初めての参加だったので脳研究に関する学会ということでとても楽しみにしていました．国際学会としては3度目の参加であったこともあり，今までより緊張せず発表することができました．初日にはeducational courseに参加させていただき，ネットワーク解析について詳しく学ぶことができました．自分が今まで知らなかった方法もたくさんあり，また，自分の英語能力不足により，理解しきれなかったところもあるので，資料を読み返し，今後の研究に活かしていきたいと思います．
脳研究の学会ということで，今まで参加させていただいた学会以上に，自分の知っている単語や手法がたくさん出てきて，自分も脳研究ができていることをすごく光栄に思いました．世界中で，こんなにも多くの脳研究が行われているということを実感することができ，もっともっと自分の研究を深めていきたいと，自分の研究に対するモチベーションもかなり上がりました．今回の発表では，結果を出し終わってから自分のミスが発覚するなど，準備がぎりぎりになってしまったので，今後もう少し余裕をもって準備できるようにしようと強く思います．発表練習などももっと念入りに行い，自分の研究のすごさをアピールできるようになりたいと思います．また，やはり英語力が乏しかったことも反省点の一つです．せっかくの機会をいただいているのに，英語力不足により議論が深められないのはもったいないので，次回参加するときには，自分の英語能力が向上したことが感じられるように，日常から英語の学習に取り組んできたいと思います．今回の反省点を今後必ず活かして，よりよい発表，研究にしていきます．
今回の学会を通して貴重な経験をさせていただくことで，世界中の研究にたくさん触れ，自分の研究，英語力向上へのモチベーションを高めることができました．この経験を活かしていけるよう日々努力をしていきたいと思います．
 

3. 聴講

今回の講演会では，下記の6件の発表を聴講しました．
 

発表タイトル       ： Basic Concepts of Network Neuroscience 著者                  ： Alex Fornito セッション名       ： An Introduction to Network Neuroscience: How to build, model, and analyse connectomes The science of complex networks, which is grounded in the mathematics of graph theory, offers a remarkably flexible and general framework for understanding brain network organization. The emerging field of network neuroscience involves the application of methods and concepts from network science to neuroscientific data. The core assumption of this approach is that the brain, like any other networked system, can be represented as a graph of nodes (representing, for example, individual neurons or neuronal populations) connected by edges (representing some measure of structural or functional interaction). Modelling nervous systems in this way allows investigators to use the rich repertoire of concepts and techniques developed in network science to understand neural organization and dynamics, providing a common language for characterizing data acquired in diverse species with different tools. Critical first steps in brain network analysis include adequately preparing data for network analysis, framing scientific questions appropriately, and understanding the strengths and limitations of different approaches for constructing brain networks. In this talk, I will discuss these issues and other core concepts of the field to provide a general introduction that covers the promise and pitfalls of network neuroscience. An understanding of these issues provides a necessary foundation for the use of more advanced topics covered throughout the workshop.

ノードとエッジの説明から，どのグラフが脳を表すのに最も適しているのかまで，幅広く脳ネットワークに関しての説明がありました．特に私たちの研究では無向グラフを主に用いているため，有向グラフについても今後学んでいきたいと思いました．また，エッジを全て同様に扱うのではない，Heterogeneous edgesという言葉も出されていました．今までそのような扱い方は知らなかったので，どんな方法があるのか自分で調べてみようと思っています．
 

発表タイトル       ： Defining Network Nodes and Extracting Timeseries 著者                  ： Janine Bijsterbosch セッション名       ： An Introduction to Network Neuroscience: How to build, model, and analyse connectomes This talk will introduce two crucial stages in any network neuroscience analysis, namely to define nodes and extract node timeseries. Node definition is achieved by parcellating the brain into a set of spatial brain regions that can be considered homogeneous functional areas (which are described by a single functional timeseries). Nodes do not typically change in their spatial layout once they are defined, so these steps are crucial for any network analysis. Concepts such as hard and soft parcellations, dimensionality, and anatomical and functional atlases will be introduced. In addition, important challenges such as between-subject differences in alignment and functional organisation will be discussed. A key focus of this talk will be to provide a critical overview of advantages and disadvantages of various alternative approaches. The aim is to provide the delegates with a comprehensive understanding of concepts and trade-offs, both from the technical methods perspective and from the applied neuroscience perspective, so that they are well equipped to make informed decisions in their future work.

ノードの定義は，様々なネットワーク解析において重要であり，私たちの研究では主に用いてきたmautomated anatomical labeling（AAL）は用いてはならないという説明がありました．今まではAAL以外を用いることまで考えたことが無かったため，重要性が分かっておらず，他の手法もあまり知りませんでした．他の方法も調べ，資料も読み返し，今後の対応を考えていきたいと思います．また，PROFUMOという初めて聞くアプローチも出てきたので，どんな手法なのか調べ，今後の応用も考えていきたいと思います．
 

発表タイトル       ：　Motivated Performance While Sleep Deprived: Reduced ACC and insula recruitment and effort-preference 著者                  ： Stijn Massar セッション名       ： ORAL SESSION: Emotion, Motivation and Social Neuroscience Sleep deprivation has profound negative impact on attentional performance (Lim 2010) and associated brain mechanisms (Ma 2015). Recent studies have begun to explore to what extent such attentional decline could be explained by loss of motivation during sleep deprivation (Liu 2016; Massar 2018). In this study we approached this issue in two ways. First, participants performed a motivated attention task once after a night of sleep deprivation and once after a night of normal sleep. Subsequently, participants performed an effort-based decision task, indicating their preference for further task performance in return for additional reward.

この発表は，睡眠不足と注意との関連について調べた研究で，被験者はそれぞれ十分な睡眠後と，一晩中起きていた後どちらのセッションにも参加し，各条件後に同じタスクを行っていました．一つ気になったのは，同じ被験者が両セッションとも行うと，睡眠との関係を調べていることが被験者にとって明確になってしまうことです．しかし，睡眠の条件を付ける時点で分かる可能性もあること，また，セッション毎に異なる被験者をとると，セッションの差に加えて個人差の影響が実験結果に影響を及ぼす可能性があることを考えると，この条件がいいのかもしれないとも思いました．このように色々な条件を考えることも今後大切にしていきたいと思います．
 

発表タイトル       ： A New Attention Node in Macaque and Human Temporal Cortex Connects to Fronto-parietal Areas 著者                  ： IIaria Sani セッション名       ： ORAL SESSION: Mapping Sensation, Perception, and Attention The cerebral cortex comprises multiple areas involved in attentional processing. Classical studies identify the parietal and frontal cortices of human and macaques as major sources of attentional signals (Corbetta et al., 2008; Kastner & Ungerleider, 2000). Recent results highlighted the existence of an area in the temporal cortex of the macaque brain that also plays a crucial role in the guidance of spatial visual attention (Stemmann et al., 2016). The discovery of this ventral attentional-control area posits new important questions about the network organization of the primate attention system and about the degree of homology between human and monkeys in the cognitive domain. We asked whether a comparative ventral attention node exists in macaques and humans, and whether it communicates via direct white matter pathways with frontal and dorsal attention areas.

私の研究でも関係のある，トップダウン注意などの言葉も出てきた，注意に関するタスクを用いた研究で，ヒトとマカクザルの比較を行っていました．どうしても注目する論文はヒトの脳のものばかりなので，他の動物と比較しているのが新鮮でした．そして，結果として，種を超えて注意の選択に重要な領域が存在するというのが興味深かったです．また，この研究だけではありませんが，3TのMRIを用いている研究が多く，世界の流れを知ることもできました．さらに他に，7TのMRIを用いている研究まで存在し，どんどん世界が進んでいることも実感できました．
 

発表タイトル       ：Shared brain connectivity patterns modulated by long-term meditation practice 著者                  ： Roberto Guidotti, Cosimo Del Gratta, Mauro Gianni Perrucci, Antonino Raffone, Vittorio Pizzella, Laura Marzetti セッション名       ： Poster Session Introduction: In the recent years meditation research has gained a great attention due to the beneficial effects on the physical and psychological status of the practitioners [1,2]. The effects of short- and long-term meditation practices on the functional and anatomical structure of the brain have also been addressed by several studies [2-4]. However, there is little knowledge on how different meditation styles can affect brain functional structure in short- and long-term meditators, more specifically on how different practices can modify the functional brain connectivity and if these modulations are shared across meditators and are depended by the meditation expertise. To address this issue, we tested whether the meditation style can be decoded using functional connectivity measured using Magnetic Resonance Imaging (fcMRI). Specifically, we investigated Focused Attention (FA) and Open Monitoring (OM) which are the main meditation styles in the Theravada practice [1], in a group of long-term Theravada monks and in a group of novice meditators. Methods: Twelve Theravada Buddhist monks (M, mean age 37.9, SD 9.4 yrs) from the Santacittarama Buddhist Monastery participated in our study. Participants practiced FA and OM meditation forms in a balanced way in this tradition. A group of ten novice meditators (M, mean age 33, SD 4 yrs) with no prior meditation experience were recruited from the local community. The participants gave their written informed consent according to the Declaration of Helsinki. The experimental procedure consisted of three blocks of the following sequence: 6 min FA and 6 min OM meditation blocks intermixed with 3 min of resting state block and cued by vocal instructions. The total duration of the experiment was 57 min. BOLD signal images were obtained using T2*-weighted echo planar (EPI) sequence (TR=4.087 s, 28 slices, voxel size 4×4×4 mm3, 860 volumes). fMRI images were temporal and motion corrected, then detrended and temporal filtered. Obtained images were further processed in order to remove white-matter, grey-matter and CSF signal, and motion confounds, finally a band-pass filter (0.01-0.1 Hz) and scrubbing was applied. Preprocessed images were divided in 14 resting state networks (RSNs) and 90 ROIs [5]. The signal within each ROI was averaged across voxels and then pairwise correlated to obtain the connectivity matrix. Connectivity matrices were calculated for each subject in each meditation block, and then the upper triangle of the matrix was extracted to fit the classifier. Linear Support Vector Machine was used to classify meditation style, separately for experts and novices. We used the 75% of the subjects for training and 25% for testing, using shuffled cross-validation to validate the model (n=200). The top 200 connections with the highest F-score were selected to reduce data dimensionality [6]. Permutation tests (PT) were used to assess for statistical significance of classification accuracy. Results: The decoding accuracy for the expert group is of 65.6% (p=0.005; PT, n=200) while for the novice group is 54.4% (p=0.07; PT, n=200). Our analysis reveals that the connectivity matrix can be used to predict the meditation style in expert monks. In the novice group, despite the above-chance accuracy, we cannot state that meditation form can be predicted by fcMRI matrices. The analysis of the most important nodes used for the decoding in the expert group shows the contribution of nodes within the Language, both posterior and anterior Salience and dorsal Default mode networks (Fig. 1). Conclusions: We showed how meditation style can be predicted by the use of fcMRI patterns, our results showed that the prediction accuracy is far better in the expert group of meditators instead of the novice group. These results suggests that long-term meditation practices strongly impact the modulation of brain networks involved in different tasks, supporting the idea that connectivity profiles could predict cognitive behavior.

この発表は，今回のOHBMで数件しか見られなかった瞑想に関する研究の一つでした．この研究では，瞑想実践者と初心者両方測定し，さらに瞑想の種類で脳状態がどう異なるのかをMRIを用いて調べた研究でした．また，サポートベクターマシーンを用いて瞑想の種類を予測した結果，初心者と比べて瞑想実践者の方が高い予測精度であり，瞑想実践が脳ネットワークの変化に影響しているとの主張でした．私も今後，今の実験設計で瞑想実践者の方も測定しようと思っていることもあり，今考えると瞑想実践者の瞑想の種類や，実践時間など，より詳細な情報が気になりました．また，Permutation testなど，私たちの研究室でも使っている人がいて，説明を聞いたことはあっても理解しきれていない処理も多かったので，これから学んで自分の研究への応用も考えていきたいと思いました．
 

発表タイトル       ：　Dynamics of large-scale Brain Network Activity at High Spatial Resolution: Methods and applications 著者                  ： Dimitri Van De Ville セッション名       ： The Pulsatile Integration at Multiple Time Scales in the Resting Brain The quest for better understanding of brain dynamics has triggered new ways to approach functional connectivity from fMRI data; i.e., using time-resolved rather than summarizing correlational measures that miss essential details of network interaction dynamics [1]. In this talk, I will highlight promising recent advances where fMRI data is analyzed at the voxel level, first, in terms of time-varying eigenvector centrality, and, second, in terms of transient activity from deconvolved BOLD signals. These new frameworks deal in different ways with spatial and temporal overlap of large-scale functional networks, and thus unravels their interdigitated and parallel organization. The first methodology captures both fine-scale and long-range interactions that can be summarized in a dynamic-FC driven parcellation, with many meaningful parcels and subdivisions [2]. The second methodology allows a frame-by-frame evaluation of transient activity and leads to innovation-driven coactivation patterns (iCAPs) [3]. The couplings (i.e., temporal overlap) between these networks provide promising new measures to build more mechanistic models of brain function at the systems level, with a large potential for interpretable disease diagnosis and prognosis [4]. References: 1) M. G. Preti, T. Bolton, D. Van De Ville. The Dynamic Functional Connectome: State-of-the-Art and Perspectives. NeuroImage, 2017, vol. 160, pp. 41-54 [DOI:10.1038/s41598-017-12993-1] 2) M. G. Preti & D. Van De Ville. Dynamics of Functional Connectivity at High Spatial Resolution Reveal Long-Range Interactions and Fine-Scale Organization. Scientific Reports, 2017, vol. 7, pp. 12773 [DOI:10.1038/s41598-017-12993-1] 3) F. I. Karahanoglu, D. Van De Ville. Transient Brain Activity Disentangles fMRI Resting-State Dynamics in Terms of Spatially and Temporally Overlapping Networks. Nature Communications, 2015, vol. 6, pp. 7751 [DOI:10.1038/NCOMMS8751] 4) F. I. Karahanoglu, D. Van De Ville. Dynamics of Large-Scale fMRI Networks: Deconstruct Brain Activity to Build Better Models of Brain Function. Current Opinion in Biomedical Engineering, 2017, vol. 3, pp. 28-36 [DOI:10.1016/j.cobme.2017.09.008]

resting stateについて，その区間のデータ全てを使ってしまうと，経時変化の情報が欠けてしまうので，動的解析を用いて解析を行っている研究でした．今後自分の研究でも動的解析を行いたいと考えているため，動的解析の詳細と共に，今まで知らなかったinnovation-driven coactivation patterns（iCAP）という同時活性化パターンの方法についても今後調べてみようと思いました．resting stateを対象とした研究が多く見られましたが，安静状態と言ってもどうしても統制が取れず，どんな状態なのかの個人差が大きいと思うので，どのように実験しているのか知っていきたいと思います．
学会参加報告書

報告者氏名	奥村駿介
発表論文タイトル	個人の全脳Parcellationが機能的結合ネットワークにおける個人差を明らかにする
発表論文英タイトル	Individual whole-brain parcellation reveals individual variability in functional connectivity
著者	奥村駿介，中村圭佑，日和悟，廣安知之
主催	Organization for Human Brain Mapping
講演会名	OHBM2019
会場	auditorium parco della musica
開催日程	2019/06/09-2019/06/13

 
 

1. 講演会の詳細

2019年06月09日から2019年06月13日にかけて，auditorium parco della musica(roma)にて開催されたOrganization for Human Brain Mapping 2019（OHBM2019）に参加いたしました．本学会は，人の脳機能に関心を寄せる神経科学者・心理学者，技術開発に関わる工学・情報学の専門家が集まり，発展の著しいニューロイメージングの知見を，広く社会で活用していく方向性について情報交換し，自分の研究に対するフィードバックを得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，学生としてM2の山本さん，古家君，大塚君，杉野さん，吉田さん，M1の風呂谷君，丹さんが参加しました．
 

2. 研究発表
   • 発表概要

私は11日のポスターセッションおよびポスターレセプションで発表を行いました．発表はポスター形式で，計2時間参加者の方と議論を行いました．
タイトルは，Individual whole-brain parcellation reveals individual variability in functional connectivityで，提案手法に基づく個人の全脳Parcellationの結果がタスクよりも個人による変動の影響を大きく受けていることを明らかにしたという内容を報告しました．以下に抄録を記載致します．

Introduction: Investigation on both intra-individual and inter-individual variability in functional connectivity networks is some of the most remarkable research being conducted at present. For example, Gratton revealed that inter-individual variation in functional connectivity is greater than intra-individual variation [1]. However, conventional research uses parcellations based on existing atlases, and analysis results strongly depend on the method of parcellation. On the other hand, it is possible to generate an individualized brain atlas segmenting the whole brain at the individual level, and enabling the generation of highly reliable atlases maintaining individual characteristics [2]. Here, we aim to examine individual variability in functional connectivity by comparing the results of individual whole brain parcellation. In this study, an individual whole brain parcellation method based on functional connectivity was proposed, and inter-individual and intra-individual variation were quantified and compared. Methods: In the proposed parcellation, the whole brain of an individual was segmented into an arbitrary number of regions using both functional and structural MRI data. First, the structural image was segmented into an arbitrary number of regions using simple linear iterative clustering (SLIC). Subsequently, a functional connectivity matrix was generated based on the segmented regions. The network defined by this matrix was segmented into multiple user-determined communities using spectral clustering. Finally, the set of regions making up each community was labeled as one region. The framework of the proposed method is shown in Fig. 1. In this study, data from 8 tasks (Sem_coh-01, Sem_coh-02, Memory_faces, Memory_scenes, Memory_words, Motor 01, Motor 02, and Rest) in 10 healthy adults randomly selected from the Midnight Scan Club were used. The number of brain parcellations was set to 25. In this experiment, individual atomic atlases with 25 regions were generated using data from 10 participants who completed 8 different tasks. Similarity was calculated between the 10 atlases and all individual atlases for the similarity matrix was embedded in two-dimensional space using multi-dimensional scaling (MDS). Subsequently, the intra-individual and inter-individual distances between tasks were calculated and compared. Results: A similarity matrix between individual atlases of 10 participants across 8 tasks (A), the distance between individual atlases visualized in two-dimensional space by MDS (B), and comparison between tasks and individual distances (C) are shown in Fig. 2. The MDS plot shows that the distribution of the atlases was small among tasks, and large among subjects. Furthermore, the average value of distances calculated between tasks and between subjects were significantly larger among subjects (p<0.05). This result reveals that the phenomenon that functional relationships are strongly influenced by individual differences rather than differences in task conditions can be reflected at the individual level. Conclusion: In this study, an individual whole brain parcellation method based on functional connectivity was proposed, and inter-individual and intra-individual variation was quantified based on the similarity between them. The results showed that distances between subjects were significantly greater than those between tasks. Therefore, it can be concluded that the result of individual parcellation using the proposed method is significantly more influenced by individual variation than by 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．本発表では質問者の名前は聞いておりません．
 
・質問内容1
Parcellationの領域数は25ROIs，50ROIs，100ROIs以外も試しているのか，またROIが増えるとMDSの分布はどう変化しているかという質問を頂きました．
この質問に対して，今回はこの3種類のみを試したと答えました．また，今回の提案Parcellationを用いた検討ではROIが増加するにつれてよりタスクよりも個人に影響を受ける結果になったと答えました．
 
・質問内容2
SLICの初期クラスタ2048分割に意味はあるのか，という質問を頂きました．
この質問に対して，今回はユーザー定義による決め打ちで決定していると答えました．
 
・質問内容3
SLICでの分割か，Spectral Clusteringでの分割かどちらが今回の結果に影響していると考えているかという質問を頂きました．
この質問に対して，構造的かつ機能的特徴を考慮してクラスタリングを行なっているSpectral clusteringの方がより影響していると答えました．
 
 
 
・質問内容4
レストデータは使用していないのかという質問を頂きました．
この質問に対して，使用した8種類のタスクの中にレストも含まれていると答えました．加えて，同じParcellation法を用いてレスト時における日間変動の検討について昨年のOHBMで発表していると答えました．
 
・質問内容5
今回の検討では，得られたParcellationの類似度行列に対してクラスタリングは試みたかという質問を頂きました．
この質問に対して，今回はクラスタリングはしていない．同一個人のタスク間距離，同一タスクの個人間距離を定量化してParcellationの分布を定量化していると答えました．
 

• 感想

近年脳機能マッピングにおいて，ParcellationやIndividual variabilityの研究に注目が集まっているためか，ポスター発表では非常に多くの方が聞きにきてくださり，密の濃いディスカッションができたように思います．今回は2度目の国際学会であったため，海外の研究者と英語で議論を行うことに対する抵抗は少なく，比較的スムーズに発表を進められたと思います．また，周りには私と同じようにParcellation方法の提案やそれを用いた解析を行っている研究者が多くおられ，脳機能マッピングにおける本研究の立ち位置を再確認することができました．本学会で得た知見とモチベーションを活かして，今後もより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4件の発表を聴講しました．
 

発表タイトル       ：Hemispheric Difference in Group, Task and Individual-dependent Variation of Functional Networks 著者                  ： Chenxi Zhao, Yaya Jiang, Xinhu Jin, Gaolang Gong セッション名       ： Poster session Abstract ： Introduction: As a topic of general interest, individual difference of human brain has been intensively studied[1,2]. By comparing human with monkeys, a previous study has shown human-specific left-lateralized anatomical variations[3], suggesting a hemisphere-dependent variation in particular brain phenotypes. Recently, the contribution of the group, task, and individual factors to the variation in whole-brain functional networks has been detangled[2]. Following this, the present study aims to examine the magnitudes of group, task and individual-dependent variations in the two hemispheric functional networks, and further evaluate the hemispheric differences. In addition, the heritability of the hemispheric differences in each type of variation was assessed. Methods: In total, 933 human connectome project (HCP) subjects (508 females,age: 22-37,212 monozygotic twins) with resting-state and task fMRI scans(emotion, language, motor, working memory) were included. All images were preprocessed by the HCP pipeline and then feed into the GRETNA toolbox[4] to do linear detrending(only resting-state fMRI), nuisance signals regression and temporally filtering (resting-state:0.01–0.1 Hz, tasks:>0.01 Hz). The AICHA atlas[5] was used to define the nodes of network(186 in each hemisphere). Functional connectivity between each within-hemispheric node pair was defined by the Pearson correlation of mean time series (z transformed). For each hemisphere, a network similarity matrix was calculated by correlating among the linearized upper triangles of hemispheric network matrices(Fig1A). As did by Gratton et al.[2], the group, task, and individual-dependent variations were calculated per subject as following: 1) the average similarity from different individuals and tasks(group, baseline), 2) the added similarity from the same task but different individuals relative to group(task), and 3) the added similarity from the same subject but different tasks relative to group(individual). The task and individual-dependent variations were compared with the group-dependent variation using paired t-tests. To test the hemispheric differences, two hemispheric variations attributable to each factor were compared using a paired t-test. For each significantly lateralized effect, the asymmetry index (AI=(L-R)/(L+R)) was calculated and its heritability (h2) was estimated using the SOLAR software[6]. Multiple comparisons were corrected by the Bonferroni method (p<0.05). Results: As shown in Fig1, both left and right hemispheric (LH and RH) functional networks showed substantial similarity across group (LH/RH:0.46±0.03/0.44±0.03) and added similarity of networks from the same individual (LH/RH:0.18±0.08/0.17±0.07), whereas subtle but significant added similarity due to task (LH/RH:0.09±0.04/0.1±0.04). Paired t-tests showed significant left-lateralized contributions of the group (t=26.8,p=0) and individual (t=4.9,p=10-6) factors to network variation, but a right-lateralized contribution of the task factor (t=20.1,p=0) (Fig2). As listed in Table1, the h2 was significant for the AI of group (h2=0.18,p=0.002) and individual-dependent variation (h2=0.24,p=7×10-5) but non-significant for the task (h2=0.18,p=0.027, not surviving the multiple-comparison correction). Conclusions: Our results demonstrated a strong hemispheric functional network stability (group-shared organization and individual features) and moderate state-dependence. Intriguingly, the variation attributable to either the group, task, or individual factors markedly differed between the two hemispheres: shared group-level factor and individual-specific features had stronger influences on the LH network organization, while state-changes had a greater impact on the RH network. Furthermore, our heritability results indicated a significant genetic role in hemispheric differences in group and individual-dependent variations, though tenuously. These findings together provide novel insight into the hemispheric functional network organization and its lateralization.

本研究は，Human Connectome Project 933人の被験者，それぞれ5種類のタスクの機能的結合ネットワークを用いて類似度解析を行いタスク間と個人間の変動比較を行っておりました．結果，個人間で最も変動が大きくなっており，Parcellationの類似度ベースで同じ検討を行っている私と全く同じ結果が得られておりました．今回私はMSCデータを用いて解析を行っておりましたが，本検討で用いられているHCPデータでもでも同じ結果が得られるか試してみたいと思いました．
 

発表タイトル       ：Influence of parcellation atlas on quality of classification of neurodegenerative diseases 著者                  ： Michaela Montilla, Martin Gajdos セッション名       ： Poster session Abstract ： Introduction: An fMRI is considered as one of potential sources providing metrics suitable for evaluation of neural changes inducted in early stages of neurodegenerative diseases and for monitoring their progress. Methods: We used graph-theory based analysis methods to classify patients within one of five groups, divided due to standard clinical diagnostic: 1. Health control (HC), 2. Alzheimer’s disease (AD), 3. Mild cognitive impairment (MCI), 4. Parkinson’s disease (PD), 5. Combination of PD and MCI. Using 7 different atlases in a segmentation step of connectivity analysis, we obtained 7 different sets of graph metrics for each subject. Overview of used atlases provides a Table 1a. Data was acquired on Siemens Prisma 3 T. A total number of 53 patients and 43 age-matched control participants with no neurological diagnosis were recruited. Demographic data on patients and healthy controls are summarized in Table 1b. T1-weighted data (1 mm iso) and functional RS-fMRI data (3 mm iso) was acquired. Data was pre-processed using SPM12. Time series were realigned, normalized into MNI152 space and filtered applying high pass temporal filter with cut off frequency 1/128 Hz. We regressed out the effect of nuisance covariates, particularly 24 movement regressors, CSF and WM regressors and spatial smoothing using a 6 mm FWHM. Overview of individual steps of processing pipeline are shown on Figure 1. After parcellation we defined single nodes as atlas ROIs and extracted representative signal as first principal component of time series. We obtained weights of edges as Pearson’s correlation coefficient and these correlation matrices were input for Brain Connectivity Toolbox functions. We used 13 different metrics to describe each of seven networks. Last step of the analysis was to classify 96 subject based on obtained metrics into one of 5 groups. We used hybrid PCA-LDA model of linear discriminative analysis and support vector machines method. Results: Classifying each subject into one of five groups with 7 different sets of metrics allowed us statistically to compare quality of classification using different definitions of regions of interest (ROIs) so the different definitions of nodes in a networks respectively. Comparing with a priory known clinic classification we evaluated quality of classification using sensitivity, specificity, accuracy and Youden index. Using one-versus-one multi class classification model, we conclude that the choice of atlas affects the success rate and thus the quality of the classification. We found out the highest values of index when classifying patients with AD (compared to the HC). Using YEO211, Harvard-Oxford and Juelich atlas we obtained specificity > 0.8 with higher values of sensitivity as well. The highest specificity value of 0.93 was achieved for the classification of PD, again compared to the control group, for the YEO211 atlas. For differentiation between patients with Alzheimer’s disease and Parkinson’s disease, the highest values of the Youden index were for atlas YEO211 and HCP. For HCP atlas, the LDA classification was achieved with both the specificity and the sensitivity values above the 0.8 limit, so the fMRI modality can be referred to as the biomarker of these diseases. Conclusions: With finding the most suitable option for brain volume segmentation, our aim was to propose the most appropriate way of determining graph properties for a given data set, so the specificity and sensitivity of classification subjects into one of neurodegenerative diseases group or health control respectively, could achieve biomarker’s values (Ritsner, 2009). By completing the classification according to the 7 atlases we found that AAL and Brodmann’s atlases used often in the neuroscience studies did not achieve the classification success at biomarker level and therefore these atlases are not as suitable for research and classification of neurodegenerative disease using the method used in this particular and similar studies as other atlases.

本研究では，7種類の異なるParcellation方法（AAL，YEO211，Craddock，Harvard-Oxford，Brodmann，Juelich，HCP MMP 1.0）を用いて，PCA-LDAモデルで症例ごとに 1. 健康対照者 (HC)，2. アルツハイマー病患者 (AD) ，3. 軽度認知障害(MCI)，4. パーキンソン病 (PD)，5. PDとMCIの混合 の5種類に分類する研究でした．本検討ではYEO211がもっとも分類に適したParcellationという結果になっておりました．集団脳Parcellationが臨床において病変の分類に適応できるというアイデアが非常に興味深かったです．
 

発表タイトル       ： A large open-source dataset of acute stroke MRIs and related automated lesion segmentation algorithm 著者                     ： Chin-Fu Liu, Sandhya Ramachandran, Victor Wang, Xin Xu, John Hsu, Susumu Mori, Michael Miller, Andreia Faria セッション名       ： Poster session Abstract ： Introduction: Stroke is one of the leading causes of death and the first cause of long-term disability. Research-wise, strokes provide a wealthy source of data for studying brain function. However, the power of a lesion study is limited by the resolution of the lesion sampling. This limit depends not only on variations in the frequency of damage across the brain but also on the multivariate pattern of damage (1). Large samples are therefore imperative for lesion-based studies, and consequently, fully automated technologies that permit large datasets to be generated and analyzed are needed. Currently, a few tools are efficient on acute stroke detection, but less accurate on lesion segmentation (2) . Others are accurate on quantification, but developed on high dimensional and relatively modest, well-characterized dataset, therefore lacking clinical generalization (3). There is a gap to be filled by large publicly available datasets associated to efficient technologies (i.e., accurate for lesion delineation, fully automated, fast, robust on clinical data, accessible to the general user) to perform stroke segmentation and quantification. We present 1) a large database of acute strokes, associated to structured radiological reports, and 2) an efficient algorithm for stroke segmentation. Methods: Large database of acute strokes: Acute stroke was defined in the MRI Diffusion Weighted Images (DWI/B0) of patients diagnosed with (first episode of) acute stroke, at the Johns Hopkins Hospital, between 2007-2017. Although other sequences (e.g., perfusion-, low resolution T1- and T2-WIs) are archived, the data presented here concerns to DWI only, as this is the most universal and eloquent sequence performed in acute stroke. This clinical DWI dataset includes 1.5 and 3T scans, diverse protocols, axial oriented, of high in plane resolution (less than 1x1mm) and high slice thickness (4-6mm). Although the technical heterogeneity, low image resolution, and clinical standards for quality control introduce noise in the image collection, it guarantees generalization to clinical scenarios. Two trained evaluators performed the manual lesion segmentation, a neuroradiologist reviewed all the cases. The process was repeated four times, until each segmentation was declared successful by consensus. Two radiologists created consensual structured radiological reports with information about type of stroke, location according different criteria (e.g., 34 brain structures and 11 vascular territories), and associated finings. Automated stroke segmentation: The pipeline (Fig. 1) involves: a) “pre-processing” steps as brain registration, skull stripping, (whitestripe) normalization and DWI scaling ; b) abnormal voxel detection , based on the “radiological normal” dataset, and properties such as brain symmetry and complementary DWI/B0 information, resulting in “ischemic” and “hemorrhagic” probabilistic maps (IS and HS); c) unsupervised deep learning (8,9) , with DWI/B0/IS/HS as inputs, trained in a subset of manual stroke delineations, in a leave-one-out manner. The accuracy of this pipeline, and its variations, is measured by the agreement with the manual delineations, in an independent subset. Results: We created a large dataset of acute strokes, with manual lesion delineation and structured reports, containing 2312 cases (401 radiological “normal”, 1512 ischemic strokes, 206 hemorrhagic, 193 mixed). Using this dataset, we created and tested models for automated lesion delineation (Fig. 2). The most efficient model (accuracy 0.95, Dice 0.9) was the “hierarchical” UnetH, which cascades a CNN and minimizes the loss function on each resolution levels of Unet. Conclusions: We generated the largest dataset of acute strokes reported so far, and an efficient algorithm for automated lesion delineation. Both will be publicly available in the near future, as a “biobank” and a web-service (10) for stroke quantification, enabling lesion-based studies and related clinical tools to achieve full potential in practice.

本研究は，急性脳卒中と診断された2312症例のMRI拡散強調画像（DWI / B0）より，自動セグメンテーション方法を用いて病変部位を検出するアルゴリズムを提案している研究でした．提案手法では，3次元のDWIの画像に対して，前処理を行った後，教師なし学習を用いて出血性脳卒中部位（HS）と虚血性脳卒中部位（IS）を検出し，Deep Learningを用いて最終的な病変部位を特定しておりました．構造画像を用いている点や教師なしでセグメンテーションを行う点が，私の研究と類似していたため，興味深かったです．教師なし学習の際に用いている特徴量に関して詳しく理解できなかったところは残念でした．
 
 

発表タイトル       ：Individual-Specific Areal-Level Parcellations Improve Functional Connectivity Prediction of Behavior 著者                  ： Ru Kong, Qing Yang, Evan Gordon, Xinian Zuo, Avram Holmes, Simon B. Eickhoff, B. T. Thomas Yeo セッション名       ： Poster session Abstract ： Introduction: The human cerebral cortex comprises hundreds of functionally distinct areas, which are in turn organized into at least ten to twenty large-scale networks. Most resting-state fMRI (rs-fMRI) parcellations have relied on group-averaged data[1–3], which might obscure individual-specific topographic features[4, 5]. Here, we propose an approach to generate individual-specific areal-level parcellations and show that the resulting parcellations can improve individual predictions of behavioral phenotypes based on functional connectivity (FC). Methods: We have previously proposed and validated a multi-session hierarchical Bayesian model (MS-HBM) for estimating individual-specific cortical networks[6]. The multiple layers of the MS-HBM explicitly differentiate intra-subject (within-subject) from inter-subject (between-subject) network variability. In contrast to MS-HBM, previous network mappings ignore intra-subject variability, so might confuse intra-subject variability for inter-subject differences. We have previously utilized MS-HBM for network parcellations[6].To estimate areal-level parcellations, the MS-HBM could be re-trained by initializing with a group-level areal-level parcellation (e.g., Schaefer2018[7]). Furthermore, we constrained the individual-specific parcels to be within 30mm of the group-level parcels, since previous studies suggest that individual variation in cortical areal location can go up to 30mm[8]. We compared MS-HBM with a well-known individual-specific parcellation approach (Gordon2017[5]). We considered rs-fMRI from 10 subjects (10 sessions each) in the MSC dataset[5]. Each subject was parcellated using all rs-fMRI sessions. Task inhomogeneity[5, 7] (standard deviation of task activation within each parcel) was then evaluated using task-fMRI data from the same subjects. A lower task inhomogeneity indicates better parcellation quality. For fair comparison, the number of MS-HBM parcels were constrained to be the same as Gordon2017. Second, we considered ICA-FIX denoised rs-fMRI data from the HCP S1200 release[9]. We selected 58 behavioral measures across cognition, personality and emotion[6]. Individual-specific MS-HBM parcellations were estimated for subjects with four runs and no missing behavior (N = 752). For each subject, we obtained 400×400 FC matrices using the 400-area group-level parcellation (Schaefer2018) or the 400-area individual-specific MS-HBM parcellations. The FC matrices were then used for predicting the 58 behavioral measures using kernel ridge regression[10]. We performed 20-fold cross-validation: kernel ridge regression was trained on 19 folds and used to predict behavior in the test fold. The regularization parameter was determined using inner-loop cross-validation. Furthermore, the 20-fold cross-validation was repeated 100 times[6]. Results: Individual-specific MS-HBM parcellations achieved better task inhomogeneity than Gordon2017, suggesting better generalization to task data (Fig. 1A). Fig. 1B shows the parcellations of two representative MSC subjects estimated from 5 rs-fMRI sessions. We observed significant topological differences between the two subjects, which were highly replicable across sessions. Compared with Schaefer2018, the FC of MS-HBM parcels achieved a higher average prediction accuracy with a relative improvement of 9.44%. We note that we could not compare with Gordon2017 because the Gordon2017 approach estimated different number of parcels in each participant, so the resulting FC matrices were not comparable across participants. Conclusions: Compared with other approaches, MS-HBM individual-specific cortical parcellations generalized better to new rs-fMRI (not shown due to space constraints) and task-fMRI data from the same subjects. MS-HBM parcellations were highly reproducible within individuals, while capturing unique individual features. Individual-specific parcellations yield better FC-based behavioral prediction compared with group-level parcellations.

本研究では，個人の全脳Parcellation方法が提案され，MSCデータセットにおけるタスクとレストのデータを用いてセッションごとの変動を検討しておりました．結果として提案Parcellationが個人の再現性が高くなることを示しておりました．Parcellation方法の提案，個人差の検討，また用いているデータが私の研究と同じであり，長時間研究の共有を行いました．
 
参考文献
https://ww5.aievolution.com/hbm1901/index.cfm?do=abs.pubSearchAbstracts
 
学会参加報告書

報告者氏名	丹真里奈
発表論文タイトル	N-back課題時のワーキングメモリ負荷量に依存した脳活動変化
発表論文英タイトル	Working memory load-dependent changes in brain activity during the N-back task
著者	丹真里奈, 日和悟, 廣安知之
主催	Organization for human brain mapping
講演会名	25th Annual Meeting of the Organization of Human Brain Mapping
会場	Auditorium Parco Della Musica
開催日程	2019/06/09-2019/06/13

 
 

1. 講演会の詳細

2019/6/9から2019/6/13にかけて，Auditorium Parco Della Musicaにて開催されましたHBMに参加いたしました．この学会は，脳画像研究やその応用に関心のある研究者や医師，学生が参加しており，発展の著しいニューロイメージングを用いた脳に関する知見を，広く社会で活用していく可能性，方向性についての情報交換，議論の場となることを目指す学会です．
本研究室からは私と廣安先生，日和先生，山本，古家，大塚，奥村(駿)，杉野，吉田，風呂谷が参加しました．
 

2. 研究発表
   • 発表概要

私は12日の12:45~14:45のセッション「Poster Session: Poster Numbers W001-W906」にて発表しました．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．
今回の発表では，「Working memory load-dependent changes in brain activity during the N-back task」と題して発表しました．以下に抄録を記載致します．

Introduction: Working memory (WM) is the ability to simultaneously hold and process information. Previous studies on the neural basis of WM examined its influence on the brain activity based on the differences observed in task performance. However, considering that the WM capacity varies among individuals, even with similar task scores, the influence of the task load on the brain activity is also considered to differ among people. Therefore, clarifying such an influence of the task load on the brain activity is of crucial importance. In the present study, the existence of a brain area which increases in activity with increasing task load was assumed. Specifically, the brain activity during the N-back task was measured by fMRI, while the brain region whose activity was altered by the changes in the WM load was extracted.   Methods: Twenty healthy adults (age: 22.4±0.17 years old, 10 females) participated in the current experiment and performed an N-back verbal identity WM task. Specifically, each participant completed three N-back fMRI runs (1, 2, 3-back) in a random order. As described in Fig. 1, each N-back run consisted of a 410-s block-design, which included both the N- and 0-back conditions in alternating 50-s blocks. Furthermore, activation analyses of the obtained fMR images were performed using SPM12 and the brain regions showing increases in activity with increasing loading amount were examined. Briefly, the activated voxels in the 3-back task with the highest load were assumed to be important for the WM load-dependent changes in brain activity. Additionally, the t-value of the 1,2,3-back task in the activated voxel at the 3-back task was extracted and the differences between the 3 groups were statistically tested. Finally, the pattern behind the WM load-dependent changes in brain activity was classified based on the differences between the 1-2-back, 2-3-back, and 1-3-back.   Results: Fig. 2(a) shows the 10 voxels activated in the 3-back task. As the brain activity increased with increasing WM load in all voxels, the regions associated with memory and attention, including the inferior parietal lobule (IPL), the superior frontal gyrus (SFG), the right precuneus (PCUN.R) and the right insula (INS.R), were examined (van der Mark et al. 2011) (Courtney et al. 1998) (Cavanna, A. E et al. 2006). The more the increase in WM load, the more the brain attentional resources were consumed. Furthermore, the precuneus also correlated with performance (p <0.001, uncorrected) and was considered to play a role in WM. Fig. 2(b) illustrates the activation levels of each of these 10 voxels at different WM loading. These areas and the changes in their activation levels were classified into three patterns; 1) pattern A: greatly increasing from the 1- to 2-back; 2) pattern B: gradually increasing as the WM load increased; and 3) pattern C: greatly increasing from the 2- to 3-back. Pattern A identified the left/right IPL, the left SFG, PCUN.R, and the left caudate nucleus (2 voxels), which are regions associated mainly to WM. Pattern B represented the INS.R and the left cerebellum, suggesting the increases in attention and discomfort to be accompanied by an increase in task difficulty. Finally, the right SFG and the left supplementary motor area reported pattern C and were greatly activated in the 3-back, which indicates their involvement in the processing of high load tasks.   Conclusions: In the current study, the WM load-dependent changes in brain activity were examined and the following brain regions, which are related to attention and memory, showed increases in activity with increases in WM load; IPL and SFG, PCUN.R, INS.R. These results suggest that the consumption of attention resources in the brain increases with increasing WM load. In addition, the existence of three patterns describing the changes in brain activity was identified in the extracted brain regions based on different activation levels seen at different WM loading amounts.

 

• 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．質問者の氏名は控え損ねてしまいました．
 
・N-back課題とは何か．
この質問に対して，私は「負荷量を変化させることができる課題．」と回答しました．
 
・なぜこれらの文字を使ったのか，大文字と小文字は同じものと判断するのか．
この質問に対して，私は「音韻ストアの機能を排除するため．大文字と小文字は違うものと判断する．」と回答しました．
 
・3-back課題時の賦活領域に着目しているが，1,2-back時にも同じボクセルが賦活したのか．
この質問に対して，私は「1,2-back時には別の領域も賦活した．また，同じ領域での賦活も確認した．」と回答しました．
 
・9ボクセルだけしか賦活しなかったのか．周りのボクセルはどうだったのか．
この質問に対して，私は「FWE-corrected，p<.001で絞ったため，閾値を変えると多くみられる．」と回答しました．
 
・GLMを用いているのか．
この質問に対して，私は「はい．」と回答しました．
 
・私はEEGを使っているが，機器は何を使用しているのか．
この質問に対して，私は「fMRI．」と回答しました．
 
・Methodの1,2-back課題時のT値抽出がよく分からない．
この質問に対して，私は「3-back課題時に賦活した領域における，1,2,3-back課題時のT値を抽出した．」と回答しました．
 
・有意差検定はANOVAを用いたのか．
この質問に対して，私は「はい．One-way ANOVAを用いた．」と回答しました．
 
・なぜ徐々に増加するということがいえるのか．1,2-backで有意差はあったのか．
この質問に対して，私は「パターンAでは1,2-backで有意差はなかった．1,3-backでのみ有意差がみられたため，徐々に増加すると考えられる．」と回答しました．
 

• 感想

初めての国際学会でのポスター発表だったため緊張しましたが，事前に準備した発表内容を思っていた以上に多くの方々に話すことができました．同じ研究テーマの方のご意見もいただくことができ，貴重な経験ができたと思います．一方で，研究テーマの方向性やポスターの作り方，英語能力の向上など，自分の今後の目標を明確にすることができました．また，ポスター発表や講演を聞き，ワーキングメモリに関する新たな知見も得られました．様々な研究分野の方が参加されていたため，自分の研究内容以外にも興味のある研究が増え，今後の研究においてとても意味のある学会参加となりました．
 

3. 聴講

今回の学会で聴講した発表のうち，下記4件を報告いたします．
 

発表タイトル       ：　Neurocognitive correlates of working memory in Postpartum Psychosis 著者                  ： Olivia S. Kowalczyk, Astrid Pauls, Montserrat Fusté, Steven Williams, Katie Hazelgrove, Costanza Vecchio, Gertrude Seneviratne, Carmine Pariante, Paola Dazzan, Mitul Mehta セッション名       ： All posters M001-M897 Abstruct            ：   Introductions: Postpartum psychosis (PP) is a severe postpartum disorder. Working memory related brain activations are consistently impaired in disorders related to PP (e.g. bipolar disorder, non-puerperal psychosis), however, few studies have investigated this in PP. The aim of this study is to compare women at risk of PP and healthy postpartum women on measures of brain activation and functional connectivity related to working memory.   Methods: Twenty-four women at risk of PP (11 developed an episode – PE; 13 remained well – NPE) and 20 healthy postpartum women completed a functional Magnetic Resonance Imaging (fMRI) scan within a year of delivery, including a classic working memory task (n-back). All fMRI data analysis was performed with FSL, following a standard preprocessing pipeline. General linear models were used to examine peak activations and psychophysiological interaction (PPI) analysis was performed to investigate task-related connectivity. For the PPI analysis the principal seeds were placed in the left and right dorsolateral prefrontal cortices (DLPFCs) based on previous studies (Owen et al., 2005). Two additional seeds (primary motor area – M1, and supplementary motor area – SMA) were used to capture connectivity associated with the motor response to the experimental conditions. Randomise (FSL) with threshold-free cluster enhancement was used for non-parametric permutations-based test of between-group differences (5000 permutations, corrected p<0.05).   Results: Hyperactivity of lateral visual areas during 0- (605 voxels), 1- (1705 voxels), and 3-back (1453 voxels) conditions was observed in the PE group compared to controls. PE and NPE women had increased connectivity with the right DLPFC and parietal, lateral visual, bilateral temporal, and cerebellar regions, during 1- (PE=1118 voxels, NPE=6847 voxels) and 2-back compared to controls (PE=48355 voxels, NPE=21826 voxels). Similarly, NPE and PE groups both had increased connectivity of bilateral parietal and visual regions with M1 compared to controls during 2-back (PE=91 voxels, NPE=16935 voxels). Additionally, the PE group had increased connectivity of the right middle temporal gyrus with the right DLPFC during 2-back compared to the NPE group (164 voxels). All p-values were <0.05.   Conclusions: This study reveals that while increased connectivity during the n-back task is evident in all women at risk of PP, there are specific increases in connectivity between the prefrontal cortex and temporal lobes in those who developed an episode. Importantly these changes differ from the reduced connectivity with the DLPFC usually observed in bipolar disorder (Cremaschi et al., 2013) and schizophrenia (Deserno et al., 2012; Quidé et al., 2013), and provide initial evidence of the potentially differential nature of abnormalities in PP. These results require replication and extension into other cognitive domains and may contribute to the development of different treatment strategies for women at risk of PP compared to those with non-puerperal psychosis.

本発表の研究は， 産後精神病のリスクがある女性と健康な女性において，ワーキングメモリに関連する脳活動と機能的結合性の比較を目的としていました．結果として，産後精神病のリスクのある女性は，N-back課題時の右DLPFCなどの結合性が増加しました．この研究では，健常群と病気のリスクがあり精神病になった群に加え，病気のリスクはあるが精神病にならなかった群の検討も行っている点が面白いと思いました．また，先行研究における統合失調症患者のDLPFCとの結合性の低下とは異なる結果について言及されており，大変興味深かったです．
 

発表タイトル       ： Network differences during an n-back working memory task in adults with autism spectrum disorder 著者                  ： Veronica Yuk, Benjamin Dunkley, Evdokia Anagnostou, Margot Taylor セッション名       ：All posters M001-M897 Abstruct            ：   Introductions: Adults with autism spectrum disorder (ASD) exhibit a varying profile of working memory (WM) abilities, though a recent meta-analysis reported that overall, adults with ASD experience some WM difficulties (Demetriou et al., 2018). These differences may be grounded in atypical functional connectivity, as individuals with ASD have shown either decreased (Urbain et al., 2016) or alternative network recruitment (Koshino et al., 2005) during WM tasks. Although it is known that WM processes are frequency-specific (Roux and Uhlhaas, 2014), and while Urbain et al. (2016) found decreased alpha-band synchronization in children with ASD, it is unclear whether this decrease continues into adulthood, or whether this difference migrates to other frequency bands with development. Given the roles of theta, alpha, and gamma oscillations in recognition (Brookes et al., 2011; Klimesch et al., 2004), we predict that adults with ASD will show decreased synchronization between WM-related regions across all three frequency bands.   Methods: We included 41 adults with ASD (27.16 ± 6.19 years; 27 males) and 38 controls (27.42 ± 6.06 years; 26 males). There were no group differences (all ps > 0.2) in age (t(77) = 0.19), sex (Χ²(1) = 9.20×10^-6), or IQ (t(74) = 1.09). Participants filled out the Behavioral Rating Inventory of Executive Function, Adult Version (BRIEF-A; Roth et al., 2005) as a measure of everyday executive function abilities; T scores on the WM subscale were compared between groups. To examine neural correlates of WM processes, participants performed a visual n-back task with abstract patterns with two loads (1 and 2) in the MEG scanner. MEG data were acquired on a CTF 151-channel system and preprocessed and analyzed in FieldTrip (Oostenveld et al., 2011). Data were epoched from -500 ms to 750 ms relative to the onset of the repeated stimulus. Source activity in the 90 AAL regions was estimated using a vector beamformer, and phase synchrony in each canonical frequency band was calculated using the weighted phase lag index. Between-group network differences were determined using Network Based Statistics (Zalesky et al., 2010), which accounts for multiple comparisons using family-wise error correction.   Results: Adults with ASD reported having more WM difficulties on the BRIEF-A (t(40.6) = 4.76, p = 2.43×10^-5), but both groups performed similarly on the n-back task (F(1,76) = 0.11, p = 0.74). Despite equivalent performance, preliminary between-group neuroimaging results in the 2-back condition showed that adults with ASD exhibit increased connectivity (t(52) = 2.5, p = 0.008) in a fronto-parietal network in the alpha band (8-14 Hz) involving the anterior cingulate cortex (ACC), the right inferior frontal gyrus (IFG), and the left superior parietal lobule (SPL). They also demonstrated decreased connectivity (t(52) = 3.0, p = 0.002) in a high-frequency gamma band (81-150 Hz) between the right superior frontal gyrus (SFG) and right IFG. An exploratory time-frequency analysis of source power in the left inferior parietal lobule (IPL) also revealed increased alpha event-related synchronization in the ASD group between 200-500 ms.   Conclusions: Given that alpha band activity during WM tasks is thought to signify inhibition of task-irrelevant information through gating mechanisms (Palva and Palva, 2011), the increased alpha band connectivity and power we observed in adults with ASD may reflect higher inhibitory control demands, which is consistent with reports that adults with ASD often show inhibition difficulties (Demetriou et al., 2018). As alpha and gamma band activity is thought to index maintenance of information in WM (Palva and Palva, 2011), decreased gamma band connectivity and increased alpha band synchronization may indicate difficulties retaining previous visual stimuli in WM for those with ASD. While this atypical connectivity did not result in accuracy differences, it may interfere with more complex behaviours involving WM, such as those experienced in day-to-day life.

本発表は， ASD患者のワーキングメモリプロセスの3つの周波数帯（シータ，アルファ，ベータ）における機能的結合性が低下することを仮説としていました．この研究では1,2-backが用いられており，それについて質問したところ，2-backが一般的であるということでした．また，N-back課題に用いている画像の種類も見たことがなくめずらしかったため，とても面白いと感じました．
 

発表タイトル       ： Task-dependent functional organizations of the visual ventral stream 著者                  ： Han-Gue Jo, Thilo Kellermann, Junji Ito, Sonja Grün, Ute Habel セッション名       ： All posters T001-T898 Abstruct            ：   Introductions: The visual ventral stream is a series of hierarchical processing stages from the primary visual cortex V1 to inferior temporal cortex IT, in which neural interactions along this hierarchy enable us to recognize visual objects. However, its complex and diverse connectivity make it difficult to illustrate the functional organization, particularly when top-down cognition is involved. Depending on task-goal, the ventral stream may require different functional structure of the hierarchy to incorporate visual features of interest into object recognition [1,2]. Here we identified context-dependent functional structures of the ventral stream.   Methods: Twenty-eight participants performed three types of visual cognition task during fMRI measurement. The three task conditions that required distinct cognitive processes for object recognition were used in order to drive the visual ventral stream: searching for a target object, memorizing objects in natural scenes, or free viewing of the same natural scenes. We identified a task-dependent connectivity network of the ventral stream, utilizing a hierarchical seed-based connectivity approach that explicitly compared task-specific BOLD time-series. Seed-based analysis was performed within the ventral stream, and the first cortical processing stage V1 was subjected as a seed region. Voxel clusters that revealed significant task effect were identified as regions of interest (ROIs) and these ROIs were further subjected as seeds for subsequent seed-based analyses. On the basis of the identified ROIs, we demonstrated task-dependent connectivity to which extent the connectivity increases or decreases during each of the visual search, memory, and free viewing conditions.   Results: The hierarchical seed-based connectivity approach identified five ROIs in the visual ventral stream (Figure 1), representing a task-dependent functional network. The connections across the identified ROIs were organized into correlated and anti-correlated structures according to the context of visual cognition. Searching for a target object separated the visual area V1 and V4 from the high-order visual area PIT (the posterior part of the IT), while memorizing objects strengthened the coupling of V4 with PIT. Furthermore, task-dependent activation was found in V1 and V4, while the PIT showed deactivation.   Conclusions: The present study demonstrated context-dependent functional structures of the visual ventral stream. In particular, while the ventral stream was organized into correlated and anti-correlated structures during searching for a target object, memorizing objects manifested a correlated structure. Our results further suggest a putative boundary between V4 and PIT, which divides the visual hierarchy into two subdivisions that interact competitively or cooperatively depending on task demand. These results highlight the context-dependent nature of the ventral stream and shed light on how the visual hierarchy is selectively mediated to bias object recognition toward features of interest.

本発表では，記憶タスクと物体の探索タスクにおける機能的結合性の違いが示唆されていました．物体の探索課題時には，後頭部の視覚関連領域における機能的ネットワークの強い増加が示されていましたが，記憶課題時には異なるネットワーク構造を示しました．ただの記憶に比べ，関心のある物体への探索では，その物体へ認識を偏らせるために，視覚的な階層がどのように仲介されるかを検討していることが大変興味深かったです．
 
 

発表タイトル       ： Cross-task Evidence for Language Recruiting an Episodic Buffer located in the Visual Word Form Area 著者                  ： Lang Qin, Bingjiang Lyu, Su Shu, Yayan Yin, Wai-Ting Siok, Jia-Hong Gao セッション名       ： ORAL SESSION(Learning and Memory) Abstruct            ：   Introductions: Language processing requires information from multidimensional perceptual and mnemonic sources to be temporarily held and collectively integrated towards intended representations, which is by definition in the charge of the episodic buffer, an important subsystem of working memory. However, neural substrates underlying the episodic buffer and its conceptual functions remain elusive. We measured functional MRI responses when subjects performing word-level spoken-/written-language comprehension and production tasks. Intriguingly, cross-task conjunction analysis identified a shared buffer located in the “visual word form area” (VWFA), namely the middle portion of left ventral occipitotemporal area (mVOT). Accordant with the buffer’s putative role, Granger causality analysis and graph analysis conjointly suggested that mVOT maintains information from multiple sources and acts as an integrative hub, under the modulation of the central executive network. These findings initially ground the theoretic buffer in neurophysiological reality and contribute to resolving the long-established VWFA debate by reconciling working memory functions and language processes.   Methods: To explore the buffer shared by multi-modal language tasks, 100 healthy adult right-handers were recruited in an fMRI experiment and instructed to participate two task sessions on separate days, where each session consisted of four single-word perception and production tasks (i.e., listening, reading, speaking and writing). Data were preprocessed with realignment, segmentation and normalization. To identify the shared buffers, conjunction analysis was conducted to locate the overlapping activatations across four tasks. Further, granger causality analysis (GCA) with model order one was applied to investigate the causal interactions between region-of-interests (ROI) for each task, with time-courses extracted from ROIs within significantly activated cortical loci. Differences between the total causal influence strength flowing into and out of a given ROI were calculated (in-degree). Functional connectivity (FC) was assessed with a brain atlas of 268 nodes covering the whole-brain (6). Having excluded the sub-cortical and cerebellar regions, the sum of weighted FC strength between a cortical locus and other loci was calculated to depict its connectivity degree, which indicates how likely it is an integrative hub (7).   Results: Cross-task conjunction analysis revealed an overlap located in the middle portion of left ventral occipitotemporal area (LvOT) (Fig. 1), which resides not only in the lexical interface of dual-stream model (8), but also in the “visual word form area (VWFA)” (9); this area has also been reported to link to semantic dementia. GCA analysis revealed significant causal information flows from multiple sources directly correlated with LvOT in all tasks with significantly high in-degrees (Fig. 2,3), indicating its critical part in holding information. Further, FC analysis found LvOT as part of the regions with strongest FC out of whole brain cortices (Fig. 4) across tasks, implying that LvOT is highly possible to act as an integrative hub. This finding is consistent with a latest research which held that middle occipitotemporal sulcus is where integration between the visual system output and the language network happens (10).   Conclusions: Firstly, our study provides compelling evidence for the existence of episodic buffer. We argue that LvOT the is an important node of the episodic buffer network, since it displayed hypothetical patterns that were generated from the putative functions of the buffer during both spoken- and written-language tasks. Also, VWFA may not be reading-specific, but rather a critical area recycled to support similar functions in reading for its special role in multi-dimensional integration (9). Lastly, this study can shed light on the “domain-general vs. domain-specific” debate by helping elucidate how language engages and specialize domain-general network.

本講演では， 言語処理において重要な役割を担うエピソードバッファの存在を明らかにするものでした．実験ではspeaking，writing，reading，listeningの4つのタスクが用いられていました．ワーキングメモリのサブシステムの1つであるエピソードバッファの機能的ネットワークとして，重要なノードの存在を主張しており，ワーキングメモリの研究をする上でとても刺激を受けました． ワーキングメモリを構成する各要素のメカニズムを研究することも重要であると感じました．
 
参考文献

• OHBM 2019-Organization for Human Brain Mapping, https://www.humanbrainmapping.org/i4a/pages/index.cfm?pageid=3882