ResUNet++: An advanced architecture for medical image segmentation

Tensoflow implementation of ResUNet++

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. More description about the archicture can be in the paper [ResUNet++: An Advanced Architecture for Medical Image Segmentation] (https://arxiv.org/pdf/1911.07067.pdf).

Architecture

Requirements:

os
numpy
cv2
tensorflow
glob
tqdm

Folders:

data: Contains the set of three dataset as mentioned.
files: Contains the csv file and weight file generated during training.
new_data: Contains two subfolder `images` and `masks`, they contains the augmented images and masks.

Files:

1. process_image.py: Augment the images and mask for the training dataset.
2. data_generator.py: Dataset generator for the keras.
3. infer.py: Run your model on test dataset and all the result are saved in the result` folder. The images are in the sequence: Image,Ground Truth Mask, Predicted Mask.
4. run.py: Train the unet.
5. unet.py: Contains the code for building the UNet architecture.
6. resunet.py: Contains the code for building the ResUNet architecture.
7. m_resunet.py: Contains the code for building the ResUNet++ architecture.
8. mertrics.py: Contains the code for dice coefficient metric and dice coefficient loss. 

First check for the correct path and the patameters.

1. python3 process_image.py - to augment training dataset.
2. python3 run.py - to train the model.
3. python3 infer.py - to test and generate the mask.

Pytorch implementation of ResUNet++ can be found here :

https://github.com/rishikksh20/ResUnet/blob/master/core/res_unet_plus.py

Dataset Link

Kvasir-SEG

Results

Qualitative results comparison on the Kvasir-SEG dataset.From the left: image (1), (2) Ground truth, (3) U-Net, (4) ResUNet, (5) ResUNet-mod, and (6) ResUNet++.

Citation

Please cite our paper if you find the work 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={Proceedings of the IEEE International Symposium on Multimedia (ISM)}, 
  title={ResUNet++: An Advanced Architecture for Medical Image Segmentation}, 
  year={2019},
  pages={225-230}}

Contact

Please contact debesh.jha@northwestern.edu for any further questions.