Towards explainable oral cancer recognition: Screening on imperfect images via Informed Deep Learning and Case-Based Reasoning

Marco Parola, Federico A. Galatolo, Gaetano La Mantia, Mario G. C. A. Cimino, Giuseppina Campisi, Olga Di Fede

Abstract: Highlights•Noise and incomplete medical data can cause bias in deep learning.•Oral cancer recognition on photos via deep learning is a viable solution for screening.•The system provides a human-centered explanation via case-based reasoning.•Ensemble at early stage of the workflow strengthens the system from artifacts and inexact annotations.•Integrating human knowledge into architectures improves the system both in terms of accuracy and visual explanation.