Code for MASS

This repository contains the official implementation of the paper "MASS: Modality-collaborative semi-supervised segmentation by exploiting cross-modal consistency from unpaired CT and MRI images".

Dataset

Our experiments were conducted on the combination of two existing datasets: BTCV and CHAOS. The combined dataset comprises 47 CT images (from BTCV) and 40 MR images (CHAOS).

To speed up data reading, we first pre-process each image using offline scaling and save it locally. In practice, you can use the script in Code/BTCV_preprocess/data_process.ipynb to pre-process the BTCV dataset, and scripts in Code/CHAOS_preprocess/process_label.ipynb and Code/CHAOS_preprocess/process_nii.ipynb to pre-process the CHAOS dataset.

Train

This is the command for using one GPU for training.

python train.py --path path_to_combined_dataset --save_path save_path --epochs N --batch_size B

Test

After finishing training, the model will be saved to save_path. Please use the following command for testing.

python test.py