HistoMapr ™ : An Explainable AI (xAI) Platform for Computational Pathology Solutions.

Pathologists are adopting whole slide images (WSIs) for diagnostic purposes. While doing so, pathologists should have all the information needed to make best diagnoses rapidly, while supervising computational pathology tools in real-time. Computational pathology has great potential for augmenting pathologists’ accuracy and efficiency, but concern exists regarding trust for ‘black-box AI’ solutions. Explainable AI (xAI) can reveal underlying reasons for its results, to promote safety, reliability, and accountability for critical tasks such as pathology diagnosis. Built on a hierarchy of computational and traditional image analysis algorithms, we present the development of our proprietary xAI software platform, HistoMapr, for pathologists to improve their efficiency and accuracy when viewing WSIs. HistoMapr and xAI represent a powerful and transparent alternative to ‘black-box’ AI. HistoMapr previews WSIs then presents key diagnostic areas first in an interactive, explainable fashion. Pathologists can access xAI features via a “Why?” button in the interface. Furthermore, two critical early application examples are presented: 1) Intelligent triaging that involves xAI estimation of difficulty for new cases to be forwarded to subspecialists or generalist pathologists; 2) Retrospective quality assurance entails detection of potential discrepancies between finalized results and xAI reviews. Finally, a prototype is presented for atypical ductal hyperplasia, a diagnostic challenge in breast pathology, where xAI descriptions were based on computational pipeline image results.