Inferring human brain structural connectivity by eigen-decomposition based network deconvolution

It is a pervasive problem that recognizing the relationship between structural connectivity detected by diffusion tensor imaging (DTI) and functional connectivity obtained from resting state functional magnetic resonance imaging (rs-fMRI) in human brain mapping. Current experimental studies indicate that functional connectivity and structural connectivity relate with each other. Furthermore, functional connection can be used to discover the links of two brain hemispheres, hard to detect with technological means such as diffusion tensor imaging, which motivates many studies to establish computational models. Here, we propose a method based on network deconvolution for inferring structural connectivity from resting-state functional connectivity, assuming that the correlation of functional connectivity are generated by the sum of one and two-step structural paths. We apply the method to a 90-ROI database and a 998-ROI database respectively, and the performance of the method is tested with ROC curves.