Single scan, subject-specific component extraction in dynamic functional connectivity using dictionary learning.

Abstract

The study of individual differences in healthy controls can provide precise descriptions of individual brain activity. Following this direction, researchers have tried to identify a subject using their functional connectivity (FC) patterns computed by functional magnetic resonance imaging (fMRI) data of the brain. Currently, there is an emerging focus on investigating the identifiability over the temporal variability of the FC. Studies have shown that dynamic FC (dFC) can also be used to identify a subject. In this study, we propose a method using the dFC and a dictionary learning (DL) algorithm to extract the subject-specific component using a single fMRI scan. We show that once the dictionary is learned using a training set, it can be stored in memory and reused for other test subjects. Using Human connectome project (HCP) and Nathan Kline Institute (NKI) datasets, we showed that our proposed method can increase the subject identification accuracy significantly from 89.19% to 99.54% using the Schaefer atlas along with subcortical nodes from the HCP atlas. The effect of monozygotic and dizygotic twins on the subject identification was also analyzed, and the results showed no significant differences between the groups having twins and the group having unrelated subjects. This proposed method can aid in the extraction of the subject-specific components of dFC.