Expert-Informed, User-Centric Explanations of Medical Image Classification

We argue that the dominant approach to explainable AI for explaining image classification with deep learning– annotating images with heatmaps, provides little value for users unfamiliar with deep learning. Instead, we argue that explainable AI for images should produce output like experts produce when communicating with one another, with apprentices, and with novices. We discuss a bit of the history of interpretable and explainable AI with examples from AI & medicine. A new approach that labels image regions with diagnostic features is proposed and evaluated. We draw on examples from radiology, ophthalmology, dermatology as well as bird classification.