Date of Award

January 2024

Document Type

Open Access Thesis

Degree Name

Medical Doctor (MD)

Department

Medicine

First Advisor

Sanjay Aneja

Abstract

Novel biomarkers can help guide management of Prostate Cancer (PCa) through the identification of high-risk phenotypes among similar patients in traditional National Comprehensive Cancer Network (NCCN) risk groups. We hypothesized that deep learning (DL) models which identified Extraprostatic Extension (EPE) and Seminal Vesicle Invasion (SVI), both pathologies associated with treatment failure, on Magnetic Resonance Imaging (MRI) could provide imaging biomarkers of PCa prognosis. In this study, two deep learning models were trained on axial T2-weighted (T2W) prostate MRI images (n=612) to derive imaging biomarkers of EPE and SVI. Area Under the Receiver Operating Characteristic Curve (AUC) was used to measure the discriminatory ability of each model on three test sets. Unsupervised hierarchal clustering of deeply learned features and GradCAM images were generated to promote interpretability. Clinical utility of EPE and SVI biomarkers was assessed with Kaplan-Meier analysis, log-rank tests were used to evaluate biochemical recurrence free survival (BrFS) for patients stratified by each biomarker, and c-indexes were calculated. Biochemical failure was defined as a post-treatment Prostate Specific Antigen (PSA) >0.1ng/mL for patients who underwent radical prostatectomy (RP) or PSA >2ng/ml above nadir for patients who received radiation treatment (RT). Within our cohort of 820 patients treated at Yale, the median age was 66.1 with a median follow up of 3.1 years. 48.4% (n=397) underwent RP and 51.6% (n=423) received RT. DL models for EPE and SVI showed good discriminatory ability, both with AUCs of 0.72. Each biomarker showed good prognostic ability to identify high risk prostate phenotypes. Patients deemed high risk based on our EPE classifier had worse 5-year BrFS (59% vs 80%, p<.001). Similarly, patients classified as high risk based on SVI also had worse 5-year BrFS (47% vs 76%, p<.001). In conclusion, deep learning classifiers of prostate MRIs demonstrated the ability to stratify high-risk prostate cancer phenotypes beyond traditional risk paradigms and imaging biomarkers represent a non-invasive method to help aid in the personalization of treatment for patients with localized prostate cancer.

Comments

This is an Open Access Thesis.

Download
Open Access

This Article is Open Access

Share

COinS