This repository provides the Pytorch code for "Segmentation of Organs at Risk Based on a Cascaded FAS-UNet+ Framework with Iterative Optimization Strategy".

Some important required packages include:

• Python == 3.8.15
• Pytorch == 1.13.1
• Some basic Python packages, such as Numpy, skimage.

1. First, you can download the test datasets at SegTHOR challenge and put them in './nii_41_60'. To save the dataset as ".npy", run:

cd ./utils/
python3 get_test_npy_data.py

2. Put pth files.

   (1). Download the 'pth' file of coarse and fine segmentation models from [BaiduPan [code:code], [BaiduPan [code:code].

   (2). Place coarse ('123_s3_fold*_1e6_avg_best.pth') and fine ('123_fold*_net2_s3_net1_avg_best.pth') segmentation models in './test/pth_net_1/' and './test/pth_net_2_by_net1/' directory, respectively.

   The files format is as follows:

    /test/pth_net_1/
        123_s3_fold1_1e6_avg_best.pth
        ...
        123_s3_fold5_1e6_avg_best.pth
    /test/pth_net_2_by_net1/
        123_fold1_net2_s3_net1_avg_best.pth
        ...
        123_fold5_net2_s3_net1_avg_best.pth

3. To test cascaded FAS-UNet+ on test dataset, run:

cd ./test/
python3 two_stage_weight_pred.py

The segmentation results are saved in the './test/2stage_result/' directory.