Transformer-based Annotation Bias-aware Medical Image Segmentation (TAB)

Project Page

This repo contains the official implementation of our paper: Transformer-based Annotation Bias-aware Medical Image Segmentation, which highlights the issue of annotator-related biases existed in medical image segmentation tasks.

Paper

Requirements

This repo was tested with Ubuntu 20.04.4 LTS, Python 3.8, PyTorch 1.7.1, and CUDA 10.1. We suggest using virtual env to configure the experimental environment.

1. Clone this repo:

git clone https://github.com/Merrical/TAB.git

2. Create experimental environment using virtual env:

virtualenv .env --python=3.8 # create
source .env/bin/activate # activate
pip install -r requirements.txt

Dataset

The dataset details and the download link can be found in the here.

Training

python main.py --dataset RIGA --phase train --net_arch TAB --masks --no_aux_loss --num_worker 8 \
--learning_rate 5e-5 --weight_decay 0.0 --num_epoch 300 \
--lambda_ 1.0 --rank 7 --loop 0

Inference

python main.py --dataset RIGA --phase test --net_arch TAB --masks --no_aux_loss --num_worker 8 \
--learning_rate 5e-5 --weight_decay 0.0 --num_epoch 300 \
--lambda_ 1.0 --rank 7 --loop 0

Bibtex

@inproceedings{Liao2023TAB,
  title={Transformer-based Annotation Bias-aware Medical Image Segmentation},
  author={Liao, Zehui and Hu, Shishuai and Xie, Yutong and Xia, Yong},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year={2023},
  organization={Springer}
}

Contact Us

If you have any questions, please contact us ( [email protected] ).