RibsegBasedonUDA

A framework for rib segmentation in CXR images based on unsupervised domain adaptation

Data

• DRRs generation: We employ a parallel projection model [1] to generate DRR images from CT images.
• Data format in Cycle-GAN: json
  • trainA
  • trainB
  • testA
  • testB
• Data format in SegNet: json
  • train
    • imgs
    • masks
  • test
    • imgs
    • masks

Usage

1. SegNet
   • cd segnet and open the file train_config.yaml and set your json path and other parameters

    python train.py train_config.yaml
   

2. Cycle-GAN
   • cd cyclegan
   • Training:

     python train.py --name yourExperName --gpu_ids 0,1 --n_epochs 100 --n_epochs_decay 100 --dataroot yourJsonDataRoot --batch_size 8

   • Test:

     cp ./log/expername/latest_net_G_A.pth ./log/expername/latest_net_G.pth
     python test.py --name yourExperName --no_dropout --dataroot yourJsonDataRoot
     

Citation

If you use this code for your research, please cite our papers.

@article{zhao2023rib,
  title={Rib Segmentation in Chest X-ray Images based on Unsupervised Domain Adaptation},
  author={Zhao, jialin and Nie, Ziwei and Shen, Jie and He, Jian and Yang, Xiaoping},
  journal={Biomedical Physics \& Engineering Express},
  year={2023}
}

References

[1] Campo, M.I., Pascau, J. and Estépar, R.S.J., 2018, April. Emphysema quantification on simulated X-rays through deep learning techniques. In 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018) (pp. 273-276). IEEE.