Low-Dimensional Embedding of Functional Connectivity Graphs for Brain State Decoding

Functional connectivity graphs are fully defined by their weighted adjacency matrix. Beyond the computation of graph-theoretical measures, we propose to use these graphs for inter-subject classification. Since they form a class of graphs with undirected edges and fixed number and ordering of vertices, vector space graph embedding techniques can be used to provide good classification performance. We propose a method that represents connectivity graphs in low-dimensional spaces, and we show experimental results hinting that such low-dimensional projections are beneficial for classification performance.