Application and development of an artificial intelligence based multiparametric malignancy index (MMI) for interpretation of 3T multiparametric prostate MRI

Introduction: Prostate cancer (PCa) diagnosis and management remain challenging due to its varied progression, from slow-growing forms to aggressive types requiring early intervention. Advances in detection, such as risk-adapted screening and multiparametric MRI (mpMRI), have improved early diagnosis, but balancing high-risk identification with minimizing overdiagnosis is crucial. This study investigates the application of machine learning (ML), specifically a Random Forest (RF) model, to enhance PCa detection accuracy using clinical and radiological features. Materials and Methods: A cohort of 314 patients aged 55-75 was analyzed, with key variables including PSA levels, PI-RADS scores, prostate volume, and rADC. The RF model was chosen for its robustness in handling mixed data types and complex feature interactions. Patients were stratified into ISUP grades (0, 1, 2+), with non-cancerous (ISUP 0) and aggressive (ISUP 2+) cases distinctly represented. A subset of 95 patients was reserved for testing to evaluate model generalization. Results: Among the cohort, 182 cases had lower PI-RADS scores (2 and 3), and 132 had higher scores (4 and 5). The RF model achieved 68% accuracy on the test set, with precision and recall for each ISUP grade: ISUP 0 (precision: 0.79, recall: 0.83), ISUP 1 (precision: 0.22, recall: 0.12), and ISUP 2+ (precision: 0.58, recall: 0.74). Feature importance analysis highlighted rADC, prostate volume, and PSA as key predictors, with the model demonstrating strong ability to distinguish between non-cancerous and aggressive PCa cases, supported by an AUC score. Conclusions: This study highlights the potential of ML to improve PCa diagnostic accuracy. The RF model’s effective integration of clinical and radiological data provides insights into key predictors, aiding in risk stratification. This work also supports the FLUTE project, which aims to enhance PCa diagnostics through federated learning, laying the groundwork for AI-driven diagnostic tools and personalized treatment planning.