Deceptive learning in histopathology
Deep learning holds immense potential for histopathology, automating tasks that are simple for expert pathologists, and revealing novel biology for tasks that were previously considered difficult or impossible to solve by eye alone. However, the extent to which the visual strategies learned by deep learning models in histopathological analysis are trustworthy or not has yet to be systematically analyzed. In this work, we address this problem and discover new limits on the histopathological tasks for which deep learning models learn trustworthy versus deceptive solutions. While tasks that have been extensively studied in the field like tumor detection are reliable and trustworthy, recent advances demonstrating the ability to learn molecular profiling from hematoxylin and eosin (H&E) stained slides do not hold up to closer scrutiny. Our analysis framework represents a new approach in understanding the capabilities of deep learning models, which should be incorporated into the computational pathologists toolkit.