by Yicheng Wu*, Zhonghua Wu, Hengcan Shi, Bjoern Picker, Winston Chong, and Jianfei Cai.

<26.07.2023> Due to IP restrictions, the data sharing is suspended now.
<11.07.2023> We release the codes.

This repository is for our MICCAI 2023 paper: 'CoactSeg: Learning from Heterogeneous Data for New Multiple Sclerosis Lesion Segmentation' (Early Acceptance, top 14%).

This repository is based on PyTorch 1.8.0, CUDA 11.1, and Python 3.8.10. All experiments in our paper were conducted on a single NVIDIA Tesla V100 GPU with an identical experimental setting.

Please obtain the original public MSSEG-2 Dataset. Then, the HD-BET tool is used to extract the brain regions. We further apply the re-sampling and z-score normalization operations here. The data split is fixed and given in 'CoactSeg/data'.

1. Clone the repository;

git clone https://github.com/ycwu1997/CoactSeg.git

2. Train the model;

sh train_mixed.sh

3. Test the model;

sh test_mixed.sh

If our model is useful for your research, please consider citing:

@inproceedings{wu2023coact,
  title={CoactSeg: Learning from Heterogeneous Data for New Multiple Sclerosis Lesion Segmentation},
  author={Wu, Yicheng and Wu, Zhonghua and Shi, Hengcan and Picker, Bjoern and Chong, Winston and Cai, Jianfei},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={3--13},
  volume={14227},
  year={2023},
  doi={https://doi.org/10.1007/978-3-031-43993-3\_1},
  organization={Springer, Cham}
  }

The current training stage is slow and there is a trick when generating the second-time-point all-lesion result on the MSSEG-2 dataset (see lines 65-66). That's because two-time-point all-lesion labels are not available for the model training and the model cannot identify such slight all-lesion differences at different time points.

We are addressing the training efficiency and the input disentanglement problems. The improved CoactSeg model will be released soon.

If any other questions, feel free to contact me at '[email protected]'

This repository is based on our previous MC-Net. We here also appreciate the public repositories of SNAC and Neuropoly, and also thanks for the efforts to collect and share the MSSEG-2 dataset and our MS-23v1 dataset from Alfred Health, Australia.