A novel automatic identification model for tracking dynamic brain functional networks at single-subject level

Recent studies based on functional magnetic resonance imaging technique suggest that the brain functional networks (BFN) may have some spatial variability over time. However, capturing the spatial dynamics of BFNs based on only single subject data is still a challenging task, mainly due to the limited samples in a sliding window. In this paper, a novel quasi group independent component analysis (quasi-GICA) model is proposed for tracking the dynamic BFNs in time effectively. The quasi-GICA consists of four main steps: firstly, the sliding window technique is used to generate multiple sliding window subsets; then, the principle component analysis is applied to project and compress each sliding window subset; thirdly, independent component analysis (ICA) is performed on the aggregated projection of all sliding window subsets, to extract the quasi-group BFNs at single subject level, embedded in whole sliding windows; finally, the multivariate regression and dual regression are proposed to extract time courses (TC) based on quasi-group BFNs from the entire data and the specifically time-varying BFNs and TCs from each sliding window subset, respectively. This quasi-GICA model theoretically overcomes some deficiencies in contrast to the existed BFNs tracking model such as lack of the effectiveness in tracking dynamic BFNs at whole brain level, additionally dynamic BFNs matching task among the multiple sliding window subsets, high time consuming, etc. And the results generated by quasi-GICA on the simulated and real data both further demonstrated the effectiveness of tracking dynamic BFNs and capturing the spatial dynamics of BFNs in time, and validated the dynamic spatial properties of human brain function. The proposed quasi-GICA model is promising to have wide application in personalized neuroimaging and neuroscience studies.