ABCD Time Series

This repository contains the code for Accounting for Temporal Variability in Functional Magnetic Resonance Imaging Improves Prediction of Intelligence.

Requirements

pytorch>1.0.0
tqdm
numpy
scipy
matplotlib
pandas

Data Preparation

Convert each individual's fMRI data into a single numpy file (shape: time_steps * # of features). fMRI data from different tasks should be put in separate folders.

|-ROOT
 |-- MID
 |-- NBACK
 |-- SST
 |-- REST
  |- NDAR_INVPG04NJDC.npy
  |- ...

Usage

Train fMRI data example

python train_fmri.py --task mid --target nihtbx_cryst_uncorrect --test_fold 0 --data_root [DATA_DIR]

Train with feature selection example

python fea_slct.py --task mid --target nihtbx_cryst_uncorrect --test_fold 0 --regcoef 0.25 --data_root [DATA_DIR]

Feature selection fine-tuning/evaluation example

python fea_slct_eval.py --task mid --target nihtbx_cryst_uncorrect --test_fold 0 --model_path [MODEL_PATH] --data_root [DATA_DIR]

Citation

If you found this code useful, please cite our paper.

@misc{abcd_time_series,
    url = {https://arxiv.org/abs/2211.07429},
    author = {Li, Yang and Ma, Xin and Sunderraman, Raj and Ji, Shihao and Kundu, Suprateek},
    title = {Accounting for Temporal Variability in Functional Magnetic Resonance Imaging Improves Prediction of Intelligence},
    year = {2022},
}