Abstract 3685: Virtual Gleason score: Towards personalized predictive identification of prostate cancer risk via integration of clinical and imaging data into spatiotemporal, biomechanistically-informed classifiers

Abstract The Gleason score (GS) is a fundamental predictor of prostate cancer (PCa) behavior and cancer-specific survival. The GS is determined by analyzing the tumor cell architecture in tissue samples. As PCa exhibits significant heterogeneity within any single tumor, whole-prostate tissue specimens enable a comprehensive assessment of the tumor’s GS. However, a biopsy, upon which treatment decisions are made, is inherently prone to sampling bias. Thus, a more accurate method to predict tumor biology and clinical disease progression is needed. To address this unmet critical demand, we propose the concept of virtual Gleason Score (vGS). The vGS consists of a personalized prediction of the local map of GS over the 3D tumor morphology extracted from MRI and informed by biomechanistic markers that characterize spatiotemporal tumor dynamics. The vGS is obtained through a three-step pipeline hybridizing physics-driven and data-driven models. First, we personalize a biomechanistic model of PCa growth using MRI and serum PSA data collected as part of routine management of the disease. Second, we calculate a panel of biomechanistic biomarker maps probing quantities and mechanisms included in the model that have shown strong correlation with GS. Finally, a machine learning algorithm leverages the local values of these biomechanistic markers to render a spatiotemporally-resolved map of vGS over the tumor geometry. Here, we present preliminary results in a cohort with 16 PCa cases that were longitudinally monitored with PSA and MRI in the context of active surveillance. The underlying biomechanistic model was the Fisher-Kolmogorov equation, and we used logistic classifiers to build the vGS. Our study shows that local proliferation activity and tumor cell density weighted by the ratio of tumor to prostate volume achieve an AUC of 0.96 as well as optimal sensitivity and specificity of 86.4% s and 90.7% for predicting GS. Despite the coarse nature of the vGS predictions obtained from the biomechanistic model used herein, these forecasts accurately anticipated the emergence of higher GS more than one year earlier than standard PCa monitoring. Future work will seek to advance the biomechanistic and data-driven models to enhance local vGS forecasts, as well as leveraging them to optimize disease management on a patient-specific basis. Citation Format: Guillermo Lorenzo Gomez, Chengyue Wu, Joshua P. Yung, John F. Ward, Hector Gomez, Thomas E. Yankeelov, Aradhana M. Venkatesan, Thomas J. Hughes. Virtual Gleason score: Towards personalized predictive identification of prostate cancer risk via integration of clinical and imaging data into spatiotemporal, biomechanistically-informed classifiers [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3685.