sd3004 / st-fmri

This repository contains code for spatio-temporal deep learning on functional MRI data

ST-fMRI

This repository contains PyTorch code for spatio-temporal deep learning on functional MRI data for phenotyping prediction. The original work was published at MLCN 2021:

Improving Phenotype Prediction using Long-Range Spatio-Temporal Dynamics of Functional Connectivity

Downloading HCP Dataset

HCP data can be directly downloaded from Human Connectome Project

Installation

For PyTorch and dependencies installation, please follow instructions in install.md

Preprocessing

In the folder /data/:

python preprocessing_nodetimeseries.py  subjects.txt 25 /data/HCP/rfMRI ../outputs/

Training Brain-MS-G3D

For sex classification

python ./tools/train_node_timeseries.py --nodes 25 --bs 64 --epochs 100 --gpu 0 --windows 100 --data_path path_to_data

For fluid intelligence regression

python ./tools/train_node_timeseries.py --nodes 25 --bs 64 --epochs 100 --gpu 0 --windows 100 --fluid --data_path path_to_data

Tensorboard

Starting tensorboard visualisation

tensorboard --logdir ./logs/MS-G3D/

Docker support

Coming soon

References

This repository is based on the following repositories:

• repository: MS-G3D - paper: Z.Liu et al 2020

• repository: ST-GCN-HCP - paper: S.Gadgil et al 2020

and

• repository: ST-GCN - paper: S.Yan et al 2018

Citation

Please cite this work if you found it useful:

@misc{dahan2021improving,
      title={Improving Phenotype Prediction using Long-Range Spatio-Temporal Dynamics of Functional Connectivity}, 
      author={Simon Dahan and Logan Z. J. Williams and Daniel Rueckert and Emma C. Robinson},
      year={2021},
      eprint={2109.03115},
      archivePrefix={arXiv},
      primaryClass={q-bio.NC}
      }