Prostate cancer ranks as the second most common cancer in men worldwide, as reported by the World Health Organization (WHO). The current diagnostic approach, which measures the prostate-specific antigen (PSA) biomarker in blood, often fails to distinguish between stable cases and those progressing to rapidly fatal tumors [1]. Despite notable advancements in simulating prostate cancer [2], a critical gap exists, necessitating the development of a model that precisely accounts for patient-specific growth dynamics. Here, we propose a data-driven physics-based model to predict tumor growth based on patient’s PSA blood test results. To achieve that, we generate a three-dimensional prostate representation from patient’s T2-weighted magnetic resonance image (MRI) sequences, incorporating cellular-level details such as cellularity, vascularization, and tumor location. We then apply a novel mathematical model based on partial differential equations (PDEs) to simulate the evolution of the PSA levels and tumor growth. Our model considers both tumor-produced PSA and its vascular transfer to the bloodstream, hence linking serum PSA dynamics to tumor progression [3]. We also incorporate a deep learning algorithm to mirror patient-specific serum PSA levels more accurately, employing a neural network within the PDE framework to estimate the proportion of proliferating tumor cells using simulation and patient data. This methodology provides a dynamic representation of tumor evolution based on PSA levels. Validation with real patient data demonstrates our methodology’s potential in forecasting prostate tumor evolution based on patient follow-up serum PSA levels.