Mohamed Omar

M.B.B.Ch

Cedars-Sinai Medical Center

Assistant Professor

PLCO (Learn more about this study)

PLCOI-1803

Feb 4, 2025

Multimodal Imaging and Pathomics for Precision Prostate Cancer Prognosis

Most patients with prostate cancer (PCa) initially present with localized indolent disease; however, a significant proportion eventually develop aggressive disease, underscoring the need for accurate and timely risk-assessment. Standard diagnostic pathways—typically involving elevated prostate-specific antigen (PSA) and transrectal ultrasound-guided biopsy—often resulted in both under-detection of clinically significant disease and overdetection of low-grade cancer. Multiparametric MRI (mpMRI) has emerged as a more precise modality for detecting clinically significant lesions and guiding targeted biopsy; however, its ability to reliably distinguish indolent from aggressive disease remains suboptimal. Similarly, clinical parameters suffer from significant prognostic limitations, with a substantial heterogeneity in outcome, especially in intermediate-risk patients. Tissue-based genomic tests have shown promise in risk-stratification; however, they have not seen universal clinical adoption. The limitations of existing approaches has often led to both undertreatment of some aggressive tumors and overtreatment of indolent disease, which carries psychological, financial, and quality-of-life burdens. Notably, existing approaches overlook the heterogeneity of the tumor microenvironment (TME), which influences tumor behavior and clinical course. We have demonstrated that PCa TME harbors heterogenous features within and across different stages, which can be leveraged for prognostication. We have also shown that embedding these features into the decision rules of prognostic models can improve their performance and generalizability.
With this in mind, our overarching goal is to develop a robust artificial intelligence (AI) risk-stratification tool for patients with PCa, that leverages standard prostate hematoxylin and eosin (H&E)-stained slides and mpMRI data. Additionally, we aim to develop methods for inferring TME signatures directly from H&E images and for integrating pathology slides with mpMRI data. These goals are based on the central hypothesis that early intratumoral dynamics leading to disease progression manifest with microscopic and macroscopic morphometric features in pathology and mpMRI data, respectively, and can be extracted using AI algorithms. These multiscale features can inform patients’ prognostication beyond what is possible with unimodal data. Our multidisciplinary team, including genitourinary radiologists, oncologists, computational biologists and AI experts, together with access to unique patient cohorts, will ensure the successful realization of the aims of this proposal.