Integrating AI in Prostate Cancer Prognosis: A Multimodal Approach

A deep learning model in a retrospective radical prostatectomy cohort more accurately distinguishes patients likely to experience biochemical recurrence from those at lower risk.

The model achieved a C-index of 0.774, exceeding conventional NCCN staging (0.706–0.746) and demonstrating superior discrimination for individual outcome ranking. This higher concordance index—derived from the same retrospective cohort using biochemical recurrence–free survival as the primary endpoint—supports better individualized risk estimates that can refine treatment selection and perioperative decisions.

Clinical variables (PSA, clinical stage, age, comorbidity where reported), histologic features (grade patterns and Gleason/ISUP grouping), and genomic signatures summarized as composite scores from targeted panels or expression-based assays were combined into a single multimodal stream.