Pneumothorax Segmentation in Routine Computed Tomography Based on Deep Neural Networks (ICoIAS'2021)

This library contains the original implementation of 'Pneumothorax Segmentation in Routine Computed Tomography Based on Deep Neural Networks' in Keras (Tensorflow as backend).

paper link: https://ieeexplore.ieee.org/abstract/document/9527604

Required libraries

• Python == 3.6
• Keras == 2.4.3
• tensorflow-gpu == 2.2.0rc3
• h5py == 2.10.0
• numpy == 1.19.4
• pydensecrf == 1.0rc3
• SimpleITK == 2.0.1
• opencv-python == 4.4.0.46

Implemented Models

• U-Net
• dilated U-Net
• ResNet34_Unet
• ResNet50_Unet
• PSPNet
• Attention U-Net
• UNet++
• MultiResUNet
• MFP-Unet
• UNet3+

Segmentation Results

Figure 1: Visualization of Pneumothorax Segmentation Results

Citation

@inproceedings{wu2021pneumothorax,
  title={Pneumothorax Segmentation in Routine Computed Tomography Based on Deep Neural Networks},
  author={Wu, Wenbin and Liu, Guanjun and Liang, Kaiyi and Zhou, Hui},
  booktitle={2021 4th International Conference on Intelligent Autonomous Systems (ICoIAS)},
  pages={78--83},
  year={2021},
  organization={IEEE}
}