MRxaI: Black-Box Explainability for Image Classifiers in a Medical Setting