Gradient-guided-Modality-Decoupling-for-Missing-Modality-Robustness

Official implementatation for paper: Gradient-guided Modality Decoupling for Missing Modality Robustness

Environment

The required libraries are listed in environment.yml

conda env create -n gmd -f environment.yml

Data preparation

download BraTS18 and modify paths in mypath.py

The structure of the dataset directory should be as follows:

<data_root>/<DATA_NAME>/*/case_name/*_flair.nii.gz      
<data_root>/<DATA_NAME>/*/case_name/*_t1.nii.gz   
<data_root>/<DATA_NAME>/*/case_name/*_t1ce.nii.gz   
<data_root>/<DATA_NAME>/*/case_name/*_flair.nii.gz
<data_root>/<DATA_NAME>/*/case_name/*_seg.nii.gz     # groundtruth 

training & eval

run sh cli/train_brats3d_gmd.sh