int_deepmreye

By : Sina KLING
Project : int_deepmreye
With : Matthias NAU, Martin SZINTE
Version: 0.1

Version description

Gaze decoder using DeepMReye network (https://github.com/DeepMReye/DeepMReye) fine tuned for experiments at MRI INT.

Installation of deepmreye

It is recommmended to work using conda virtual environments. Install DeepMReye with a CPU/GPU version of TensorFlow using the following command.

conda create --name deepmreye python=3.9
conda activate deepmreye
pip install deepmreye
git clone git@github.com:sinaklg/int_deepmreye.git
cd int_deepmreye
pip install -r requirements.txt

For GPU support (if available and nvidia), install GPU toolkits:

conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0

Model Weights and Example Data

Model weights and example data are available here.

Download the example dataset and model weigths by running:

python download_example.py int_dataset_example

The example contains 3 participants BOLD timeseries preprocessed and registered to individual T1w structural scans using fmriprep 23.1.4.

Data repository will be structured as such:

int_dataset_example/DeepMReyeCalib/derivatives/int_deepmreye
├── figures
├── func
│   ├── sub-0X
│   │   ├── sub-0X_ses-00_task-project_name_run-01_space-T1w_desc-preproc_bold.nii.gz
│   │   ├── sub-0X_ses-00_task-project_name_run-02_space-T1w_desc-preproc_bold.nii.gz
├── log_outputs
│   ├── deepmreye_gpu014_8969452_4294967294.err
│   ├── deepmreye_gpu014_8969452_4294967294.out
├── mask
│   ├── sub-0X
│   │   ├── mask_sub-0X_ses-00_task-project_name_run-01_space-T1w_desc-preproc_bold.p
│   │   ├── mask_sub-0X_ses-00_task-project_name_run-02_space-T1w_desc-preproc_bold.p
│   └── sub-0Y
│       ├── mask_sub-0Y_ses-00_task-project_name_run-01_space-T1w_desc-preproc_bold.p
│       ├── mask_sub-0Y_ses-00_task-project_name_run-02_space-T1w_desc-preproc_bold.p
├── model
│    └── modelinference_DeepMReyeCalib.h5
├── pp_data
│   ├── sub-0X_project_name_no_label.npz
│   ├── sub-0Y_project_name_no_label.npz
├── pred
│   ├── evaluation_project_name.npy
│   ├── scores_project_name.npy
│   ├── sub-0X
│   │   ├── sub-0X_pred_median.tsv.gz
│   │   ├── sub-0Y_pred_subTR.tsv.gz
├── report
│   ├── sub-0X
│   │   ├── report_sub-0X_ses-00_task-project_name_run-01_space-T1w_desc-preproc_bold.html

To run

All analaysis parameters are in settings.json. Make sure to specify this file to your experiment settings beforehand. Just run the main script with correct input (see header in decode_gaze.py

python decode_gaze.py [main_dir] [project_dir] [task]

Example :

python download_example.py int_dataset_example
python decode_gaze.py int_dataset_example DeepMReyeCalib