Artificial intelligence faces reproducibility crisis as unpublished code and sensitivity to training conditions make many claims hard to verify. This is also the case for AI in medicine. For example, for the field of computational pathology, despite an ever-growing number of publications, only few methods are reused by other researchers and even fewer have entered a clinical routine workflow. A team of Helmholtz Munich researchers now analyzed how to improve reusability and reproducibility of these deep learning algorithms.