MM-GTUNets

This repository is the official PyTorch implementation of "MM-GTUNets: Unified Multi-Modal Graph Deep Learning for Brain Disorders Prediction".

Contents

1. Installation
2. Configuration
3. Data
4. Training
5. Testing
6. Citation

Installation

Code developed and tested in Python 3.9.0 using PyTorch 2.0.0. Please refer to their official websites for installation and setup.

Some major requirements are given below:

numpy~=1.26.2
scikit-learn~=1.2.2
scipy~=1.10.1
torch~=2.0.0
torch-geometric~=2.0.4
nilearn~=0.10.1

Alternatively, you can choose to run the following code to install the required environment:

pip install -r requirements.txt

Configuration

Please see Configs >>>here Lines 44-96<<<.

Configs	Custom key	note
dataset	ABIDE/ADHD-200	dataset
seed	random seeds	default is 911
train	train or test (int)	1-train, 0-test
fold	10 (int)	k-fold validation
early_stop	early stop patience (int)	default is 100
lr	initial model learning rate (float)	default is 1e-4
vae_lr	initial vae learning rate (float)	default is 1e-3
epoch	number of epochs for training (int)	default is 500

other args: * --node_dim dimension of node features after modality alignment * --img_depth depth of the img_unet * --ph_depth depth of the ph_unet * --hidden hidden channels of the unet * --out out channels of the unet * --dropout ratio of dropout * --edge_drop ratio of edge dropout * --pool_ratios pooling ratio to be used in the Graph_Unet * --smh graph_loss_smooth * --deg graph_loss_degree * --val graph_loss_value

Data

ABIDE

To fetch ABIDE public datasets.

python fetch_abide.py

ADHD-200

The pre-processed ADHD-200 data upload address is as follows:

Google Drive

Link：https://drive.google.com/drive/folders/19HoajzuBFIV0dVGLtWv_jx2c0qg9srX_?usp=sharing

Baidu Cloud Drive

Link：https://pan.baidu.com/s/16sqz0fZvuSHHypMkLtikbA

Password：qj12

Training

Classification Task ( Default dataset is ABIDE )

python train_mm_gtunets.py --train 1

Testing

Classification Task ( Default dataset is ABIDE )

python train_mm_gtunets.py --train 0

Citation

If you find our codes helpful, please star our project and cite our following papers:

@article{cai2025mm,
  title={MM-GTUNets: Unified multi-modal graph deep learning for brain disorders prediction},
  author={Cai, Luhui and Zeng, Weiming and Chen, Hongyu and Zhang, Hua and Li, Yueyang and Feng, Yu and Yan, Hongjie and Bian, Lingbin and Siok, Wai Ting and Wang, Nizhuan},
  journal={IEEE Transactions on Medical Imaging},
  year={2025},
  publisher={IEEE}
}