dFCExperts

This repository is the official implementation of paper "dFCExperts: Learning Dynamic Functional Connectivity Patterns with Modularity and State Experts".

Dataset

The fMRI data used for the experiments of the paper should be downloaded from the Human Connectome Project and ABCD_ABCC.

Example structure of the data folder

data (specified by option --sourcedir)
├─── hcp1200
│    ├─── label.csv
│    ├─── hcp_rest_datasplit_5folds.pth
│    ├─── hcp_rfMRI_REST1_LR_fc_Schaefer2018_400Parcels.pt
│    └─── hcp_rfMRI_REST1_LR_tc_Schaefer2018_400Parcels.pt
├─── abcd_abcc
│    ├─── label.csv
│    ├─── hcp_rest_datasplit_5folds.pth
│    └─── hcp_rfMRI_REST1_LR_tc_Schaefer2018_400Parcels.pt
└─── samples
     ├─── sample_timeseries_data.pth
     ├─── sample_split_6folds.pth
     └─── label.csv

Requirements

To install requirements:

pip install -r requirements.txt

Training

To train the model(s) with given sample data, run this command:

python3 main.py --exp_name 'hcp_c' \
                --dataset 'hcp-sample' \
                --targetdir './result' \
                --target_feature 'Gender' \
                --gin_type 'moe_gin' \
                --num_gin_experts 5 \
                --num_states 7 \
                --state_ex_loss_coeff 10 \
                --orthogonal \
                --freeze_center \
                --project_assignment \
                --fc_hidden 256 \
                --num_epochs 30 \
                --minibatch_size 8 \
                --train \
                --validate \
                --test \
                --test_model_name 'model_val_acc'

Acknowledgements

Parts of the implementation in dFCExpert are adapted from the STAGIN repository, developed by Byung-Hoon Kim et al.

We thank the authors of STAGIN for making their code publicly available.
The adapted portions are used in accordance with the STAGIN license, which is included in this repository as LICENSE-STAGIN.txt.

If you use this repository in your work, please also cite the original STAGIN paper:

Kim B H, Ye J C, Kim J J. Learning dynamic graph representation of brain connectome with spatio-temporal attention[J]. Advances in Neural Information Processing Systems, 2021, 34: 4314-4327. [https://proceedings.neurips.cc/paper_files/paper/2021/file/22785dd2577be2ce28ef79febe80db10-Paper.pdf]