CaraNet: Context Axial Reverse Attention Network for Small Medical Objects Segmentation

This repository contains the implementation of a novel attention based network (CaraNet) to segment the polyp (CVC-T, CVC-ClinicDB, CVC-ColonDB, ETIS and Kvasir) and brain tumor (BraTS). The CaraNet show great overall segmentation performance (mean dice) on polyp and brain tumor, but also show great performance on small medical objects (small polyps and brain tumors) segmentation.

🔥 NEWS 🔥 The full paper is available: CaraNet

The journal version is available: CaraNet

Architecture of CaraNet

Backbone

We use Res2Net as our backbone.

Context module

We choose our CFP module as context module, and choose the dilation rate is 8. For the details of CFP module you can find here: CFPNet. The architecture of CFP module as shown in following figure:

Axial Reverse Attention

As shown in architecture of CaraNet, the Axial Reverse Attention (A-RA) module contains two routes: 1) Reverse attention; 2) Axial-attention (The code of axial attention is applied from UACANET)

Installation & Usage

Enviroment

• Enviroment: Python 3.6;
• Install some packages:

conda install pytorch==1.1.0 torchvision==0.3.0 cudatoolkit=10.0 -c pytorch

conda install opencv-python pillow numpy matplotlib

• Clone this repository

git clone https://github.com/AngeLouCN/CaraNet

Training

• Download the training and testing dataset from this link: Experiment Dataset
• Change the --train_path & --test_path in Train.py
• Run Train.py
• Testing dataset is ordered as follow:

|-- TestDataset
|   |-- CVC-300
|   |   |-- images
|   |   |-- masks
|   |-- CVC-ClinicDB
|   |   |-- images
|   |   |-- masks
|   |-- CVC-ColonDB
|   |   |-- images
|   |   |-- masks
|   |-- ETIS-LaribPolypDB
|   |   |-- images
|   |   |-- masks
|   |-- Kvasir
|       |-- images
|       |-- masks

Testing

• Change the data_path in Test.py

Evaluation

• Change the image_root and gt_root in eval_Kvasir.py
• You can also run the matlab code in eval fold, it contains other four measurement metrics results.
• You can download the segmentation maps of CaraNet from this link: CaraNet
• dice_average.m is to compute the averaged dice values according to sizes of objects, for small area analysis.

Segmentation Results

• Polyp Segmentation Results

• Conditions of test datasets:

• Small polyp analysis

The x-axis is the proportion size (%) of polyp; y-axis is the average mean dice coefficient.

Brain Tumor Segmentation

• Dataset

BraTS input	Segmentation truth

• Results

• Small tumor analysis

For very small areas (<1%):

The difference between results of CaraNet and PraNet:

Citation

If you think our work is helpful, please cite both conference and journal version.

@inproceedings{lou2021caranet,
author = {Ange Lou and Shuyue Guan and Hanseok Ko and Murray H. Loew},
title = {{CaraNet: context axial reverse attention network for segmentation of small medical objects}},
volume = {12032},
booktitle = {Medical Imaging 2022: Image Processing},
organization = {International Society for Optics and Photonics},
publisher = {SPIE},
pages = {81 -- 92},
year = {2022},
doi = {10.1117/12.2611802}}

@inproceedings{9506485,
  author={Lou, Ange and Loew, Murray},
  booktitle={2021 IEEE International Conference on Image Processing (ICIP)}, 
  title={CFPNET: Channel-Wise Feature Pyramid For Real-Time Semantic Segmentation}, 
  year={2021},
  volume={},
  number={},
  pages={1894-1898},
  doi={10.1109/ICIP42928.2021.9506485}}
  
@article{lou2023caranet,
  title={CaraNet: context axial reverse attention network for segmentation of small medical objects},
  author={Lou, Ange and Guan, Shuyue and Loew, Murray},
  journal={Journal of Medical Imaging},
  volume={10},
  number={1},
  pages={014005},
  year={2023},
  publisher={SPIE}
}