Resting-state brain dynamic modes predict behavioral traits

Poster No:

1375 

Submission Type:

Abstract Submission 

Authors:

Shigeyuki Ikeda1,2, Koki Kawano1, Soichi Watanabe1, Okito Yamashita1,2, Yoshinobu Kawahara1,3

Institutions:

1RIKEN Center for Advanced Intelligence Project, Tokyo, Japan, 2ATR Neural Information Analysis Laboratories, Kyoto, Japan, 3Institute of Mathematics for Industry, Kyushu University, Fukuoka, Japan

First Author:

Shigeyuki Ikeda  
RIKEN Center for Advanced Intelligence Project|ATR Neural Information Analysis Laboratories
Tokyo, Japan|Kyoto, Japan

Co-Author(s):

Koki Kawano  
RIKEN Center for Advanced Intelligence Project
Tokyo, Japan
Soichi Watanabe  
RIKEN Center for Advanced Intelligence Project
Tokyo, Japan
Okito Yamashita  
RIKEN Center for Advanced Intelligence Project|ATR Neural Information Analysis Laboratories
Tokyo, Japan|Kyoto, Japan
Yoshinobu Kawahara  
RIKEN Center for Advanced Intelligence Project|Institute of Mathematics for Industry, Kyushu University
Tokyo, Japan|Fukuoka, Japan

Introduction:

Dynamic functional connectivity in resting-state functional magnetic resonance imaging (rs-fMRI) has gained much attention because of its relations with behavioral traits. This study used dynamic mode decomposition (DMD) to extract dynamic modes (DMs), spatial-temporal coherent patterns, inherent in resting-state brain activity. To validate the effectiveness of DMD, we investigated whether individual differences in various behavioral traits were predicted using DMs and multivariate pattern analysis. Furthermore, we confirmed whether DMD outperformed ICA, a conventional method that focuses on either space or time.

Methods:

We used publicly available rs-fMRI and behavioral data of the Human Connectome Project (HCP) 1200-subjects release [7]. After a consultation with Safety Management Division in RIKEN, we confirmed that we did not need our institutional approval in the use of the HCP dataset. This study included 829 healthy young adult subjects (men 388 and 441 women, mean age: 28.6 ± 3.7 years). Individual subjects underwent four rs-fMRI runs of 14 min and 33 s each (1200 frames per run), with eyes open with relaxed fixation on a projected bright cross-hair on a dark background. Our analysis began with the rs-fMRI data which was preprocessed based on the HCP preprocessing pipelines. 59 behavioral measures were selected from the HCP dataset based on a previous study [3], which covered seven behavioral categories, e.g., Cognition. To remove confounding factors, age, gender, race, education, and head motion averaged across all the runs were regressed out from each behavioral measure.
Each rs-fMRI image was parcellated into 268 nodes using a previous functional atlas [1,6] and rs-fMRI time series for each node were calculated by averaging voxel-wise time series within each node, resulting in a 268 × 1200 data matrix for each run. To extract spatial-temporal coherent patterns inherent in resting-state brain activity, each data matrix was decomposed into DMs using DMD [4,5]. Each of DMs is described by an eigenvector (i.e., coherent spatial pattern) and a corresponding eigenvalue (i.e., frequency and growth/decay). For comparison, each data matrix was decomposed into independent components using temporal ICA.
As input features for prediction, we constructed a gram matrix representing similarity between subjects using eigenvectors of DMs. However, because there is no natural correspondence between DMs of different subjects, we cannot calculate similarity between vectors composed of eigenvectors using a conventional method such as cosine similarity. To overcome the problem, we used the Grassmann manifold, on which a subspace spanned by eigenvectors of each subject can be represented as a point [2]. The similarity between two points on the Grassmann manifold is computed using the Frobenius norm. Consequently, a gram matrix (829 × 829) was created using data for each run, and a mean gram matrix was computed by averaging gram matrices of all the runs.
To predict each behavioral trait, we used Gaussian process regression and a leave-one-family-out cross-validation procedure. The prediction accuracy was estimated by calculating the Pearson correlation coefficient (r) between predicted and true results. A p value corresponding to the prediction accuracy was computed by a permutation test (10,000 permutations). The statistical significance of the prediction accuracy was tested by applying a false discovery rate correction (q ≤ 0.05) to 59 behavioral measures.

Results:

As shown by Fig. 1, DMD outperformed temporal ICA. Specifically, we observed significant prediction results in seven of 59 behavioral measures, e.g., Dimensional Change Card Sort (r = 0.17). Most of the significant behavioral measures fell into the Cognition category.
Supporting Image: Fig1_ed.jpg
   ·Fig. 1 Prediction results.
 

Conclusions:

We showed that DMs computed by DMD explained individual differences better than spatial patterns computed by temporal ICA. Our results suggest that DMs are primarily associated with cognitive traits.

Higher Cognitive Functions:

Higher Cognitive Functions Other

Modeling and Analysis Methods:

Classification and Predictive Modeling 2
fMRI Connectivity and Network Modeling 1
Multivariate Approaches

Keywords:

Cognition
Computational Neuroscience
Data analysis
FUNCTIONAL MRI
Machine Learning
Multivariate
NORMAL HUMAN

1|2Indicates the priority used for review

My abstract is being submitted as a Software Demonstration.

No

Please indicate below if your study was a "resting state" or "task-activation” study.

Resting state

Healthy subjects only or patients (note that patient studies may also involve healthy subjects):

Healthy subjects

Was any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.

Not applicable

Was any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.

Not applicable

Please indicate which methods were used in your research:

Functional MRI
Behavior
Computational modeling

For human MRI, what field strength scanner do you use?

3.0T

Which processing packages did you use for your study?

Other, Please list  -   HCP preprocessing pipelines

Provide references using author date format

[1] Finn, E. S. (2015) ‘Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity’, Nature neuroscience, 18(11), pp. 1664–1671.
[2] Hamm, J. (2008) ‘Grassmann discriminant analysis: A unifying view on subspace-based learning’, in Proceedings of the 25th international conference on Machine learning, pp. 376–383.
[3] Liégeois, R. (2019) ‘Resting brain dynamics at different timescales capture distinct aspects of human behavior’, Nature communications, 10(1). doi: 10.1038/s41467-019-10317-7.
[4] Rowley, C. W. (2009) ‘Spectral analysis of nonlinear flows’, Journal of fluid mechanics, 641(1), pp. 115–127.
[5] Schmid, P. J. (2010) ‘Dynamic mode decomposition of numerical and experimental data’, Journal of fluid mechanics, 656, pp. 5–28.
[6] Shen, X. (2013) ‘Groupwise whole-brain parcellation from resting-state fMRI data for network node identification’, NeuroImage, 82, pp. 403–415.
[7] Van Essen, D. C. (2013) ‘The WU-Minn Human Connectome Project: An overview’, NeuroImage, 80, pp. 62–79.