Core codes of Pytorch implementation of MEDIA'20 paper and an improved version of MICCAI 19.
Need to specify the path of data label and image features Data labels are in csv format. Image features can be saved in npz format with clustering label, etc. More can be found in dataset definition.
For example, patient 10000 has 100 sampled patches from at least 1 WSI, should stored as 10000.npz with the following information:
More can be found in dataset definition.
A csv file in the following formate is needed: patient_ID | Img_patch_path | Survival_time | Survival_status |
---|---|---|---|
10000 | /10000/1.jpg | 1000 | 0 |
Our implementation consists in a main.py file from which are imported the MIL dataloader definition MIL_dataloader.py, the model architecture DeepAttnMISL_model.py and some miscellaneous training utilities.
After specific your label and feature path, run:
python main.py
The average C-index across 5 folds will show in the end.
If you find this repository useful in your research, please cite:
@article{yao2020whole,
title={Whole Slide Images based Cancer Survival Prediction using Attention Guided Deep Multiple Instance Learning Networks},
author={Yao, Jiawen and Zhu, Xinliang and Jonnagaddala, Jitendra and Hawkins, Nicholas and Huang, Junzhou},
journal={Medical Image Analysis},
volume={65},
pages={101789},
year={2020},
publisher={Elsevier}
}
@inproceedings{yao_deep_2019,
author = {Yao, Jiawen and Zhu, Xinliang and Huang, Junzhou},
booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
copyright = {All rights reserved},
pages = {496--504},
publisher = {Springer},
title = {Deep Multi-instance Learning for Survival Prediction from Whole Slide Images},
year = {2019}
}