Dynamic weighted "small-world" graphical network establishment for fNIRS time-varying brain function analysis

Abstract With the increasing application of functional near-infrared spectroscopy (fNIRS) technology, topological brain network analysis has recently been developed and successfully applied in fNIRS-based brain research. However, the current network information is analyzed through the binary network, and it lacks a dynamic estimation. In this study, we proposed a novel dynamic weighted “small-world” topology network and applied it to fNIRS analysis. Firstly, we introduced a novel flat-top (FT) window to estimate sliding-window correlation (SWC), improving the spectrum performances. And then, we applied the least absolute shrinkage and selection operator (LASSO) algorithm to the time-varying correlation matrix and obtained corresponding sparse matrix results. Finally, we established a dynamic weighted topological graph and calculated “small-world” network parameters to analyze the brain network dynamics. Simulation results showed that our proposed FT-based SWC method realized wider bandwidth, better spectrum performance, and more accurate dynamic tracking capability with the minimum mean square errors(MSEs) results in each iterative simulation. The fNIRS results showed that the average node degree (DE) and global efficiency (GE) reached a peak in the middle and late periods of the task. The network parameters showed similar changes in 2-back and 3-back tasks, which differ from the 0-back task. Statistical analysis revealed the DE and GE of the subnetwork decreased as the task became more difficult. These results revealed the typical time-varying characteristics of the brain network dynamics during the working memory process, which were expected to unveil the brain mechanisms underlying the cognitive neural activity in greater depth.