Patient Selection for Active Surveillance for Prostate Cancer based on Deep Features from U-Found: An MRI-based Foundation Model of the Prostate

Motivation: Patients with low-risk prostate cancer are often put under active surveillance (AS) to delay treatment until cancer progression. Selecting correctly is crucial to disease management. Goal(s): The aim is to determine if the embeddings of U-Found can be utilized to assist in the clinical decision making whether a patient is at high risk for progression. Approach: Using multiparametric (mp)MRI data from contemporary clinical trial, we obtained embeddings from Apparent Diffusion Coefficient (ADC) maps. We developed a model using features related to progression. Results: The prediction model reached AUC of 97%, indicating strong cancer signal available in the embeddings. Impact: We present an application of novel MRI-based foundation model of the prostate to assess prostate cancer progression risk for AS patients, thereby providing essential data for clinicians that can prospectively improve AS patient selection.