Multiview Cortical Graph Isomorphism Network (MC-GIN) for MCI/NC classification based on PET/MR sync data

Getting Started

The whole project consists two parts: Clinica Surface Processing and Graph Neural Network construction (VGAE and GIN).

##Model

Installing

The conda environment is as follows

dgl=0.9.0
nibabel=4.0.1
pytorch=1.10.2
scikit-learn=1.0.2
scipy=1.7.3

Quick distribution:

conda install --yes --file requirements.txt

Results

Use area under the ROC curve (AUC) and average precision (AP) scores for each model on the test set. Numbers show mean results and standard error for 10 runs with random initializations on fixed dataset splits.

Method	ACC	SEN	SPE	AUC
SVM
GCN
GAT
g-GIN
l-GIN
MC-GIN

Hyperparameter

特征种类、脑区数量

Built With

• Clinica - The web framework used
• DGL - Main framework

Authors

• Tianli Tao @ShanghaiTech University - Initial work