README.md

ATPGCN: Adversarially-Trained Persistent Homology-Based Graph Convolutional Network for Disease Identification Using Brain Connectivity

A preliminary version of ATPGCN with demo data which is different from ones in our paper. The example is just used to replicate our framework.

Setup

Sparse Brain Network

We first construct the functional brain connectivity with an open multimodal interface. The method also integrates the ROI-wise group constraint for regularization.

Generate_BrainNet_01.py

Adversarial Example Generation

If it is desired to generate the brain connectome perturbations and perform the adversarial training (optional).

Generate_Prbs_02.py

Persistent homology-Based Topology Feature

We extract the persistent homology features of brain conectivity network from an algebraic topology analysis.

Generate_TopoFeat_03.py

Model Training and Testing

Main_04.py