Cross-View Mutual Learning for Semi-Supervised Medical Image Segmentation (ACM'MM 2024 Poster)

Authors: Song Wu, Xiaoyu Wei, Xinyue Chen, Yazhou Ren, Jing He, Xiaorong Pu.

Official code and datas for "Cross-View Mutual Learning for Semi-Supervised Medical Image Segmentation". (ACM'MM 2024)

1. Requirements

This repository is based on PyTorch 1.9.1, CUDA 11.6 and Python 3.9.15. All experiments in our paper were conducted on an NVIDIA GeForce RTX 3090 GPU with an identical experimental setting. You should pip install some packages for reproducing our experiments:

scikit-image
scipy
tensorboardX
nibabel
medpy
h5py
numpy==1.23 (the version >1.24 may cause conflicts with medpy)

2. Workflow of CML

Overall pipeline of CML in the co-training framework. We first apply the CutMix operation to modify the inputs and supervisory signals to conduct the supervision objective $L_{sup}$. Specifically, for labeled data, we conduct the same CutMix to generate mixed labels $Y^l_{mix}$, which are used to supervise the predictions of $X^l_{mix}$. For unlabeled data, we construct heterogeneous supervisory signals $Y^u_{mix,0}$ and $Y^u_{mix,1}$, obtained by the combination of two subnets' pseudo-labels, to supervise the predictions of $X^u_{mix}$. Note that $L_{dis}$ is a conflict-based unsupervised loss, aiming to learn distinct features from the same input.

3. Usage

We provide code, and data for LA and ACDC datasets.

To train a model,

python CML_LA_train.py
python CML_ACDC_train.py

To test your trained model, and get the final performance,

python test_LA.py
python test_ACDC.py

4. Datasets

Data could be got at LA and ACDC.

Particularly, we provide the complete LA and ACDC datasets in cloud with key: data. You can download directly, and move them to the folder data.

5. Acknowledgements

Our code is modified from URPC, SS-Net and BCP. Thanks to these authors for their valuable work.

6. Citation

If you use our code or datas in this repository for your research, please cite our papers.

@inproceedings{ACMMM24CML,
  author = {Song Wu, Xiaoyu Wei, Xinyue Chen, Yazhou Ren, Jing. He, Xiaorong Pu},
  title = {Cross-View Mutual Learning for Semi-Supervised Medical Image Segmentation},
  journal ={ACM Multimedia (ACM MM)},
  year = {2024},
}

If you have any problems, please contact me by songwu.work@outlook.com.