An AI-driven Framework for Evaluating Prostate Cancer Progression

Prostate cancer progression is a complex and heterogeneous process that cannot be fully captured by deterministic models or by reliance on a single biomarker such as prostate-specific antigen (PSA). While PSA is widely used in clinical practice, it provides an incomplete and sometimes misleading representation of the underlying tumor dynamics, particularly in cases of low PSA but significant disease burden or during treatment response. In this study, we utilize anonymous clinical data from prostate cancer patients and propose a stochastic modeling framework to characterize the temporal evolution of prostate cancer as a complex process, incorporating both deterministic biological mechanisms and stochastic variability across patients. The proposed model introduces a latent disease state representing tumor burden, which evolves according to drift and diffusion components reflecting tumor growth, treatment effects, and intrinsic biological uncertainty. In addition to PSA, key clinical variables such as Gleason grade group, disease stage, and treatment exposure are integrated into the model to enhance its clinical interpretability and predictive capability. Numerical simulations demonstrate that the stochastic framework captures clinically meaningful behaviors, including heterogeneous progression trajectories, treatment-induced declines in PSA, and divergence between observed PSA levels and true disease burden. Unlike traditional survival or regression-based approaches, the model provides a full probability distribution of disease states over time, allowing for uncertainty quantification and personalized risk assessment. The results suggest that incorporating latent-state stochastic dynamics can significantly improve the understanding and prediction of prostate cancer progression, offering a foundation for next-generation decision-support systems in precision oncology.