Figure 3.

Summary of the results: figure plots the best detection rates achieved on all the training and test sets using five commonly used network features implemented in the BCT, our method and our method without the high power voxel selection step. Features 1–5 are the degree, topological overlap, clustering coefficient, local efficiency and rich club coefficient respectively. (A,B) Show the results on the training sets when the classification is performed on all the subjects and on the male and female subjects separately. (C,D) Show the similar results on test sets. The detection rates of (B,D) are computed by averaging the detection rates on the male and female groups.