Explainable Radiomics-Based ML for Predicting Clinically Significant Prostate Cancer in Biparametric MRI

Motivation: Prostate MRI interpretation is complex and subjective, facing challenges in diagnosis Prostate Cancer (PCa). Interpretable AI models are needed to aid diagnostic process. Goal(s): To develop explainable Machine Learning (ML) models using MRI-derived radiomic features to classify clinically significant prostate cancer (csPCa). Approach: A retrospective study using MRI exams of 344 patients was conducted. Radiomic features were extracted from T2-weighted and Apparent diffusion coefficient maps. ML classifiers were trained and evaluated, with SHapley Additive exPlanations (SHAP) analysis used for interpretability. Results: Random Forest model achieved an AUC of 0.85, with SHAP analysis clarifying radiomic feature contributions and enhancing model transparency. Impact: The development of explainable ML models using biparametric MRI radiomic features enhances csPCa classification, proposing a framework that connects prediction and interpretability. This approach can lead further research into transparent AI tools, benefiting clinical decision-making.