Computer-aided methods are nowadays an integral part of the medical diagnosis process. They support radiologists and other medical doctors with functionalities such as segmentation, registration, or quantification. Expert systems for instance, exploit extensive knowledge bases, to support medical doctors. However, these systems are not as broadly accepted as envisioned by their developers, because the medical expert has to curate the knowledge base, which is a time-consuming task. Modern deep learning techniques on the other hand, exploit large scale training data sets instead of curated knowledge bases. These systems promise to enable new breakthroughs, which have not been imagined before, due to the high accuracies obtained. Unfortunately, learning-based systems come with their own set of challenges, ranging from data availability to interpretability.
During this spring school, we would like to explore how we can combine rule-based and learning-based approaches in medical AI, and thus not only obtain accurate predictions but also consider interpretability and ethical implications. Therefore, we will explore challenges and solutions related to the four main pillars of medical AI: