Explainable methods for medical imaging datasets.
Can be installed dusing pip Dockerfile and docker-compose is provided too.
To process the datasets, see the notebooks process_DDSM, process_IAD and process_HAM10000 Code to train attribute predictors is in train_abl Code to train linear classifiers on top of either human, resnet or human+resnet features is provided in train_classfier.py Code to train the joint model is provided in train_joint.py Code to train the autoencoders is in train_VAE.py
Code is provided as a guide, it will not necessarily work as is because of the data dependencies but documentation and implementation details should be useufl to replicate our results in your data.