FDR-controlled network modeling and analysis of fMRI and sEMG signals

Neural recording technologies such as functional magnetic resonance imaging (fMRI) and surface electroencephalography (sEMG) provide great potential to studying the underlying neural systems and the related diseases. A broad range of statistical methods have been developed to model interactions between neural components. In this thesis, a false discovery rate (FDR)-controlled exploratory group modeling approach is introduced to model interaction/cooperation between neural components. Group network modeling for comparison between populations is of great common interest in biomedical signal processing, particularly when there might be considerable heterogeneity within one or more groups, such as disease populations. A group-level network modeling process, the group PCfdr algorithm with taking into account inter-subject variances, is proposed. The group PCfdr algorithm combines group inference with a graphical modeling approach for discovering statistically significant structure connectivity. Simulation results demonstrate that the group PCfdr algorithm can accurately recover the underlying group network structures and robustly control the FDR at user-specified levels. To further extract informative features and compare the connectivity patterns across groups at the network level, network analysis methods including graph theoretical analysis, lesion and perturbation analysis are applied to examine the inferred networks. It can provide great potential to investigate the connectivity patterns as well as the particular changes associated with certain disease states. The proposed network modeling and analysis approach is applied to fMRI data collected from control and Parkinson’s Disease (PD) groups. The network analysis results of the PD groups before and after L-dopa medication support the hypothesis