6. Moumen Elmelegy, PhD

Prostate cancer (PCa) is the second most prevalent cancer among men worldwide. The main challenge is to increase PCa early detection ensuring that high-risk patients are identified and treated appropriately while minimizing overdiagnosis and unnecessary interventions (e.g., biopsies). The latest NCCN Guidelines (March, 2024) for PCa early detection rely exclusively upon PSA tests and DREs. The guidelines then consider mpMRI or other biomarkers, including blood-based (e.g., prostate health index and 4Kscore) and urine-based (e.g., SelectMDx, ExoDx, MyProstateScore). However, PSA screening remains controversial due to its low specificity for clinically significant PCa. DRE is uncomfortable for patients and often inconclusive. Although mpMRI and blood/urine-based biomarkers with genetic analysis may offer improved diagnostic performance, they are often limited by availability, cost, and technical expertise, particularly in less-developed regions.

This research generally explores the potential of a cost-effective, non-invasive AI-powered tool that leverages the combination of PSA-derived biomarkers, risk factors (age, family history, race), and key symptoms (such as urinary issues) to assess a man’s risk of clinically significant prostate cancer (Gleason score ≥ 7). By combining these factors, the tool aims to provide a calibrated score to guide decision-making or the need for further diagnostic tests like MRI or biopsy.

Data from a cohort of 84 patients of median age 65±7.7 years (45 benign prostatic hyperplasia, 39 prostatic carcinoma) was collected from Mansoura University Hospital in Egypt. The dataset included patient’s age, PSA level, and a personalized symptom assessment on a 0-5 scale. Ground truth diagnoses were established via biopsy-based Gleason score. A comprehensive suite of state-of-the-art machine learning algorithms, encompassing both tree-based and deep learning methods, was employed to develop a clinically significant PCa predictive model. The results (sensitivity=0.78, specificity=0.88, AUC= 0.80) indicate that the predictive model shows great promise as a cost-effective PCa risk prediction tool.