Regression activation mapping on the cortical surface using graph convolutional networks

Ben A Duffy, Mengting Liu, Trevor Flynn, Arthur W Toga, A. James Barkovich, Duan Xu, Hosung Kim

Abstract: Saliency mapping on graph convolutional networks (GCNs) is important for model interpretability and has not yet been investigated for regression GCNs employing graph downsampling. We examined regression activation mapping for localizing the salient regions identified by GCNs. We first demonstrated using simulations that it is possible to generate precise vertex-wise saliency maps on the cortical surface mesh and then applied this method to age prediction using cortical surfaces derived from neonatal structural MRI.

Code Of Conduct: I have read and accept the code of conduct.

Remove If Rejected: Remove submission from public view if paper is rejected.