ROJS-Net: Multi-objective Joint Segmentation Network for Tumor and Organs-at-risk

This is the project web for the study titled "Multi-objective Joint Segmentation Network for Tumor and Organs-at-risk".

Installation

Install PyTorch and torchvision from http://pytorch.org and other dependencies. You can install all the dependencies by

pip install -r requirements.txt

Data Preparation

The folder structure of the data should be like

data/
  ├── index
    ├── train_path_list.txt
    ├── val_path_list.txt
    ├── test_path_list.txt
  ├── TrainingImage
    ├── image_1.nii.gz
    ├── image_2.nii.gz
    ├── ...
  ├── TrainingMask
    ├── organ_1.nii.gz
    ├── organ_2.nii.gz
    ├── ...
  ├── TrainingTumor
    ├── tumor_1.nii.gz
    ├── tumor_2.nii.gz
    ├── ...

Pre-training

To pre-train our ROJS-Net, run train.py. The weights will be saved in ./result/res_semmoe_prompt/. You can also use the pre-trained checkpoints of ROJS-Net in the ./result/res_semmoe_prompt/.

Test

Run predict.py, and the segmented image will be saved in ./result/res_semmoe_prompt/prediction/, then can obtain the Dice, HD95 and ASD values by running compute_value.py.

Citation

If this code is helpful for your study, please cite:

4. References