A two-step clustering-based pipeline for big dynamic functional network connectivity data.

Abstract

Dynamic functional network connectivity (dFNC) estimated from resting-state functional magnetic imaging (rs-fMRI) studies the temporal properties of FNC among brain networks by putting them into distinct states using the clustering method. The computational cost of clustering dFNCs has become a significant practical barrier given the availability of enormous neuroimaging datasets. To this end, we developed a new dFNC pipeline to analyze large dFNC data without accessing hug processing capacity. We validated our proposed pipeline and compared it with the standard one using a publicly available dataset. We found that both standard and iSparse kmeans generate similar dFNC states while our approach is 27 times faster than the traditional method in finding the optimum number of clusters and creating better clustering quality.