ngohgia/brain-surf-cnn

Reference

Gia H. Ngo, Meenakshi Khosla, Keith Jamison, Amy Kuceyeski, Mert R. Sabuncu. From Connectomic to Task-evoked Fingerprints: Individualized Prediction of Task Contrasts fromResting-state Functional Connectivity (Accepted to MICCAI 2020)

[Project Webpage]
[Paper]

Overview

This project contains the source code for BrainSurfCNN, a surface-based convolutional neural net-work to predict individual task contrasts fromtheir resting-state fingerprints. The model was adapted from the UGSCNN model github.

1. data folder contains the surface mesh templates, medial-wall mask and subject IDs from the Human Connectome Project (HCP) S1200 used in our experiments for MICCAI2020 paper.
2. experiments folder contains the scripts to replicate our MICCAI 2020 experiments, and a test example to try out the code on a small sample dataset.
3. model folder contains BrainSurfCNN source code.
4. preprocess folder contains functions needed for preprocessing the surface data.
5. utils folder contains utility functions to run the experiment and perform post-hoc evaluation.

Getting Started

1. Set up conda environment with the environment.yml file provided:

conda env create -f environment.yml
source activate brain_surf_cnn

Please note that the Pytorch version used in our experiments were installed with CUDA 10.2, but the code is compatible with CUDA 9 as well.

2. Download HCP Workbench https://www.humanconnectome.org/software/get-connectome-workbench for data preprocessing.

3. Download HCP S1200 and HCP Retest dataset https://db.humanconnectome.org/, which are used in our MICCAI 2020 experiments. In particular, the preprocessed resting-state timeseries and individual subject's task-based z-stats contrast maps are needed. Such files can be found in the following relative paths on HCP AWS S3 repository:

$SUBJ/MNINonLinear/Results/$SESSION/${SESSION}_Atlas_MSMAll_hp2000_clean.dtseries.nii
$SUBJ/MNINonLinear/Results/tfMRI_$TASK/tfMRI_${TASK}_hp200_s2_level2_MSMAll.feat/GrayordinatesStats/cope${COPEID}.feat/zstat1.dtseries.nii

where $SUBJ is the subject's ID, $SESSION is the resting-state sessions (rfMRI_REST1_LR, rfMRI_REST1_RL, rfMRI_REST2_LR, rfMRI_REST2_RL), $TASK is the 7 fMRI tasks (LANGUAGE, RELATIONAL, SOCIAL, EMOTION, WM, MOTOR, GAMBLING), $COPEID is the COPE IDs of the specific task contrasts. The exact contrasts (and COPE IDs) used in our experiments can be found in preprocess/3_separate_task_cifti.sh.

4. HCP1200 Parcellation+Timeseries+Netmats (PTN) https://db.humanconnectome.org/data/projects/HCP_1200 data are also needed for computing the resting-state fingerprints.

5. Run data preprocessing with the scripts under preprocess folder.

6. Run training and prediction with the scripts under experiments folder.

Bugs and Questions

Please contact Gia at ngohoanggia@gmail.com