Model-Heterogeneous Semi-Supervised Federated Learning for Medical Image Segmentation

Introduction

This is an official release of the paper Model-Heterogeneous Semi-Supervised Federated Learning for Medical Image Segmentation, including the network implementation and the training scripts.

Model-Heterogeneous Semi-Supervised Federated Learning for Medical Image Segmentation,
Yuxi Ma, Jiacheng Wang, Jing Yang, Liansheng Wang
Published in: IEEE Transactions on Medical Imaging (TMI)

Code List

• Network
• Training Codes
• Models Weights

Usage

Dataset

In this paper, we perform the experiments using two imaging modalities, including the polyp images (Kvasir, CVC-ClinicDB, CVC-ColonDB, CVC-300, EndoTectETIS), and ISIC-2018 images.

Training

Run the train script $ python train.py.

Testing

Please download the pre-trained weights from Baidu Disk (https://pan.baidu.com/s/1MHkadVGC3UVCVchR2AtAVg?pwd=8m7y, 8m7y) and put them in the project directory.

Rename the log_dir as ./checkpoints.

Run the test script $ python test.py.

Result

The test Dice scores and HD95 scores on the Polyp dataset are:

Citation

If you find HSSF useful in your research, please consider citing:

@ARTICLE{Ma2023Model,
  author={Ma, Yuxi and Wang, Jiacheng and Yang, Jing and Wang, Liansheng},
  journal={IEEE Transactions on Medical Imaging}, 
  title={Model-Heterogeneous Semi-Supervised Federated Learning for Medical Image Segmentation}, 
  year={2023},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/TMI.2023.3348982}}