Explainable artificial intelligence in prostate cancer treatment recommendation: A decision support system for oncological expert panels

Abstract Multidisciplinary cancer conferences (MCC) provide an expert platform for evaluating and selecting the best possible oncological treatment options. Artificial intelligence (AI) can be a highly promising technology to complement additional treatment recommendations (TR). In this context, we developed an AI system supporting prostate cancer (PC) TRs. Data from PC patients receiving MCC recommendations (2015-2022) were converted into a machine-readable format to train classifiers replicating TRs using machine learning and deep learning techniques. A two-step process identified superordinate recommendation categories (high-level) before specifying detailed TR (low-level). AI training was performed on 5478 MCC cases (76 patient input and 23 output parameters). The AI system generated fully automated TR with excellent F1-scores for high-level (e.g. surgery (0.89), radiation therapy (0.81)) and low-level (e.g. prostatectomy (0.99), PSMA-Ligand therapy (0.98)). Explainability is provided by clinical features and their importance score. To our knowledge, this study presents the first AI-generated explainable TR in metastatic and non-metastatic PC with multi-target and feature space, achieving excellent performance results.