Fig. 1: Strategy for estimating brain network states from resting state fMRI.

BOLD images were clustered into brain states according to their spatial patterns, i.e., images with similar spatial patterns were clustered together. Specifically, the preprocessed fMRI data of each individual subject was projected to the FreeSurfer fsaverage3 surface space, which consisted of 1284 vertices in total. The data of different time points of each subject were aligned to construct a spatiotemporal matrix (indicated by the red rectangle). The data of all subjects were then concatenated and temporally standardized. Principal component analysis (PCA) was applied to the concatenated data matrix to reduce the dimensionality of data to 650, which could explain more than 99% of the variance of the original data. The k-means clustering algorithm was then performed to classify the fMRI frames into a certain number of clusters (e.g., 19 clusters in this study). The fMRI frames assigned to the same cluster were averaged to generate the brain state map.