Community Detection in Multi-Aspect Functional Brain Networks: Robust Tensor Decomposition Approach

Functional organization of the brain can be characterized as a network of interconnected regions. Study of brain networks has offered new insights on human behavior and neurodegenerative diseases. Recent studies show that the functional connectivity networks are dynamic with the topology and community structure evolving with time during task performance and rest. Most of the current work on analyzing the community structure of dynamic networks either focus on single subjects or extract the group level community structure. In this paper, we present a framework for community detection in dynamic functional connectivity networks across multiple subjects and at the individual level, simultaneously. The proposed approach is based on a structured robust tensor decomposition with spectral clustering, temporal smoothness and co-clustering regularization terms to extract both the group and individual level community structures. Moreover, the temporal evolution of the community structure within the group is tracked to identify change points that may correspond to events of interest. The proposed framework is applied to dynamic functional connectivity networks constructed from task-based electroencephalogram (EEG) data.