Brainnetome4Depression

1. Environment and Data

Environment

conda create --name bn4depression python=3.9.1
source activate bn4depression

conda install -c conda-forge nibabel
conda install -c conda-forge nilearn
conda install matplotlib
conda install yaml
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia
pip install tqdm

Data

adjust your own path in load_path.py Dataset in OpenNeuro: depression_ds002748

.
├─Brainnetome4Depression
│  └─BN_Atlas
├─connection_matrix
└─depression_ds002748
    ├─sub-01
    │  ├─anat
    │  └─func
    ├─ ... ...

2. Run on Windows

Get functional connection: python functional_connection.py
Change your hyperparameter and save_model_weights/save_result_txt in config.yaml
Change your aggregation type in run.py and then python run.py

3. Run on BSCC Platform

module load anaconda/2021.11 
module load cudnn/8.8.1_cuda11.x 
# create bn4depression via Conda
chmod -x run.sh
source activate bn4depression
dsub  -s run.sh #提交作业
djob # 查看作业id
djob  -T 作业ID #取消作业

4. Original Data

result_lobe1/2/3.txt: aggregation type is lobe.
result_gyrus1/2/3.txt: aggregation type is gyrus.
To save public store, all txt files were zipped as ori_data.zip.
draw.ipynb: AUC and LogLoss with lobe/gyrus or different epochs/learning_rate.