You have accessJournal of UrologyCME1 Apr 2023PD22-02 HISTOPATHOLOGY-INFORMED RADIOLOGY BIOMARKERS IMPROVE ARTIFICIAL INTELLIGENCE-BASED DETECTION OF AGGRESSIVE PROSTATE CANCER ON MAGNETIC RESONANCE IMAGING Indrani Bhattacharya, Karin Stacke, Richard Fan, James Brooks, Mirabela Rusu, and Geoffrey Sonn Indrani BhattacharyaIndrani Bhattacharya More articles by this author , Karin StackeKarin Stacke More articles by this author , Richard FanRichard Fan More articles by this author , James BrooksJames Brooks More articles by this author , Mirabela RusuMirabela Rusu More articles by this author , and Geoffrey SonnGeoffrey Sonn More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003295.02AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Artificial Intelligence (AI) methods for aggressive prostate cancer detection on Magnetic Resonance Imaging (MRI) can help standardize radiologist interpretations. However, existing methods are often inaccurate, partly due to the use of radiology features alone without considering histopathology. Histopathology images contain definitive information about the presence and aggressiveness of cancer. Identifying correlations between radiology and histopathology images of the same tissue enables understanding the radiologic appearance of cancer and the discovery of histopathology-informed radiology biomarkers that emphasize cancer features, improving aggressive prostate cancer detection. METHODS: We developed an AI system that uses correlation learning to identify histopathology-informed MRI biomarkers using MRI and histopathology images of patients who underwent radical prostatectomy. The learned histopathology-informed MRI biomarkers are then used by the AI system to selectively identify and localize aggressive and indolent prostate cancer on MRI. Once trained, these MRI biomarkers can be extracted in new patients without pathology, aiding clinical diagnosis. We trained the system using 75 patients who underwent radical prostatectomy, and 24 patients with normal prostates (no cancer on biopsy). We evaluated the system on a lesion level using 40 patients who underwent radical prostatectomy. True positives were assessed using ground truth lesion outlines, while false positives were assessed using prostate sextants without cancer. RESULTS: Our proposed system improved aggressive prostate cancer detection over a baseline method without histopathology-informed MRI biomarkers (ROC-AUC: 0.90±0.22 vs. 0.86±0.28), correctly detecting 94% of aggressive cancers, while correctly predicting 51% of negative sextants as cancer-free. The system had fewer false positives and better overlap with cancer labels than the baseline method. CONCLUSIONS: MRI biomarkers that correlate with histopathology images emphasize aggressive cancer features and improve detection on MRI. These biomarkers can be identified from radiology and histopathology images, and do not need histopathology images during testing, making them clinically useful. Source of Funding: Departments of Radiology and Urology, Stanford University, National Cancer Institute of the National Institutes of Health (R37CA260346 to M.R, and U01CA196387, to J.D.B.), and the generous philanthropic support of our patients (G.S.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health © 2023 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 209Issue Supplement 4April 2023Page: e664 Advertisement Copyright & Permissions© 2023 by American Urological Association Education and Research, Inc.MetricsAuthor Information Indrani Bhattacharya More articles by this author Karin Stacke More articles by this author Richard Fan More articles by this author James Brooks More articles by this author Mirabela Rusu More articles by this author Geoffrey Sonn More articles by this author Expand All Advertisement PDF downloadLoading ...
Read full abstract