Deep neural networks (DNN) have been successfully applied to various prediction tasks in rs-fMRI, but the feature selection mechanism of it often appear to be a black box. We developed understanding of DNN’s prediction mechanism and proposed a feature selection method based on each feature’s contribution to the prediction. Experiments were done on the functional connectivity (FC) gender prediction to extract gender related brain FC patterns with 1003 subjects’ rs-fMRI data. The proposed method was validated by the cross-entropy loss of each feature’s prediction, and results showed the selected features are robust and consistent with the findings in previous studies.
This abstract and the presentation materials are available to members only; a login is required.