Implementation of "Modal Uncertainty Estimation for Medical Image Diagnosis".

This is a pytorch reimplementation of the MUE model proposed in this paper.

To train the model, have a look at the functions in /pipeline/train_lib.py and configuration examples provided in /pipeline/configs. It should straightforward to create a train_main.py file (preferably in /pipeline) using the functions therein and your own associated .gin configuration files. Then run the following command from the project root directory

python -m PATH_TO_TRAIN_MAIN

to train the model.

For evaluation, refer to the functions defined in /pipeline/eval_lib.py, and create your eval_main.py file similarly.
Example for loading the evaluation results and compute the Generalized Energy Distance metric can be found in plot_LIDC.ipynb.

If you find this research helpful, consider citing our paper:

@incollection{qiu2021modal,
  title={Modal Uncertainty Estimation for Medical Imaging Based Diagnosis},
  author={Qiu, Di and Lui, Lok Ming},
  booktitle={Uncertainty for Safe Utilization of Machine Learning in Medical Imaging, and Perinatal Imaging, Placental and Preterm Image Analysis},
  pages={3--13},
  year={2021},
  publisher={Springer}
}