UM-CAM

This repository provides a weakly supervised method utilizing image-level labels for medical segmentation.

If you use this toolkit, please cite the following paper:

• J. Fu, T. Lu, S. Zhang, G. Wang, UM-CAM: Uncertainty-weighted multi-resolution class activation maps for weakly-supervised fetal brain segmentation, in: MICCAI, 2023, pp. 315–324.

BibTeX entry:

@inproceedings{fu2023cam,
title={UM-CAM: Uncertainty-weighted Multi-resolution Class Activation Maps for Weakly-supervised Fetal Brain Segmentation},
author={Fu, Jia and Lu, Tao and Zhang, Shaoting and Wang, Guotai},
booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
pages={315--324},
year={2023},
organization={Springer}
}

How to use

1. Raw Psuedo label based on UM-CAM

1.1 Train the classification network

python code_cls/train_cls.py --root_path data_dir

1.2 Generate CAMs

python code_cls/generate_cam.py --root_path data_dir --layer_num 30

1.3 UM-CAM

python code_cls/generate_umcam.py

2. Refined pseudo label based on GSE method

python code_cls/GSE_refinement.py

3. Noise-robust learning based on RVC strategy

python code_seg/train_RVC.py

Requirements

Before you can use this package for image segmentation. You should:

• PyTorch version >=1.12.1
• Some common python package such as Numpy, Pandas, OpenCV, scipy, SimpleITK, ......
• Intall the FastGeodis for geodesic distance transformation

Acknowledgement and Statement

• We thank the authors of pytorch-grad-cam, FastGeodis, MIDeepSeg and PyMIC for their elegant and efficient code base
• This project was designed for academic research, not for clinical or commercial use.