Skip to content

Official implementation of ResUNet++, CRF, and TTA for segmentation of medical images (IEEE JBIHI)

Notifications You must be signed in to change notification settings

DebeshJha/ResUNetPlusPlus-with-CRF-and-TTA

Repository files navigation

ResUNet++-with-Conditional-Random-Field-and-Test-Time-Augmentation

This is the extension of our previous version of the ResUNet++. In this paper, we describe how the ResUNet++ architecture can be extended by applying Conditional Random Field (CRF) and Test-Time Augmentation (TTA) to further improve its prediction performance on segmented polyps. The GitHub code for the ResUNet++ can be found at here.

ResUNet++

The ResUNet++ architecture is based on the Deep Residual U-Net (ResUNet), which is an architecture that uses the strength of deep residual learning and U-Net. The proposed ResUNet++ architecture takes advantage of the residual blocks, the squeeze and excitation block, ASPP, and the attention block.

ResUNet++: An Advanced Architcture for Medical Image Segmentation

Architecture

Datasets:

The following datasets are used in this experiment:

  1. Kvasir-SEG
  2. CVC-ClinicDB
  3. CVC-ColonDB
  4. ETIS-Larib polyp DB
  5. ASU-Mayo Clinic Colonoscopy Video (c) Database
  6. CVC-VideoClinicDB

Hyperparameters:

  1. Batch size = 16
  2. Number of epoch = 300
  3. Loss = Binary crossentropy
  4. Optimizer = Nadam
  5. Learning Rate = 1e-5 (Adjusted for some experiments)

Results

Qualitative result comparison of the proposed models with UNet, ResUNet, and ResUNet++ on Kvasir-SEG dataset

Qualitative result comparison of the model trained on CVC-612 and tested on Kvasir-SEG

Qualitative result comparison of the model trained on CVC-612 and tested on Kvasir-SEG

ROC curve of the model trained on Kvasir-SEG dataset

Citation

Please cite our work if you find it useful.

@INPROCEEDINGS{8959021,
  author={D. {Jha} and P. H. {Smedsrud} and M. A. {Riegler} and D. {Johansen} and T. D. {Lange} and P. {Halvorsen} and H. {D. Johansen}},
  booktitle={2019 IEEE International Symposium on Multimedia (ISM)}, 
  title={ResUNet++: An Advanced Architecture for Medical Image Segmentation}, 
  year={2019},
  pages={225-255}}
@article{jha2021comprehensive,
  title={A comprehensive study on colorectal polyp segmentation with ResUNet++, conditional random field and test-time augmentation},
  author={Jha, Debesh and Smedsrud, Pia Helen and Johansen, Dag and de Lange, Thomas and Johansen, Havard and Halvorsen, Pal and Riegler, Michael},
  journal={IEEE Journal of Biomedical and Health Informatics},
  year={2021},
  publisher={IEEE}
  

}

Contact

Please contact [email protected] for any further questions.