Mp07-18 Evaluation of a Biparametric Mri AI Algorithm for the Detection of Prostate Cancer Using Spatial Annotations on Wholemount Prostate Pathology

You have accessJournal of UrologySurgical Technology & Simulation: Artificial Intelligence I (MP07)1 May 2024MP07-18 EVALUATION OF A BIPARAMETRIC MRI AI ALGORITHM FOR THE DETECTION OF PROSTATE CANCER USING SPATIAL ANNOTATIONS ON WHOLEMOUNT PROSTATE PATHOLOGY Charles Hesswani, Enis C. Yilmaz, Stephanie A. Harmon, David G. Gelikman, Christopher R. Koller, Sahil H. Parikh, Kyle C. Schuppe, William S. Azar, Daniel Nethala, Neil Mendhiratta, Alexander P. Kenigsberg, Sandeep Gurram, Baris Turkbey, and Peter A. Pinto Charles HesswaniCharles Hesswani , Enis C. YilmazEnis C. Yilmaz , Stephanie A. HarmonStephanie A. Harmon , David G. GelikmanDavid G. Gelikman , Christopher R. KollerChristopher R. Koller , Sahil H. ParikhSahil H. Parikh , Kyle C. SchuppeKyle C. Schuppe , William S. AzarWilliam S. Azar , Daniel NethalaDaniel Nethala , Neil MendhirattaNeil Mendhiratta , Alexander P. KenigsbergAlexander P. Kenigsberg , Sandeep GurramSandeep Gurram , Baris TurkbeyBaris Turkbey , and Peter A. PintoPeter A. Pinto View All Author Informationhttps://doi.org/10.1097/01.JU.0001008728.41882.d7.18AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Advancements in artificial intelligence (AI) have improved its performance in detecting and interpreting suspicious prostate cancer (PCa) lesions on biparametric magnetic resonance imaging (bpMRI). Validation of AI performance compared to pathology at radical prostatectomy (RP) is challenging due to scarcity of wholemount spatial annotations used to train AI models. This study evaluates the ability of our in-house automated deep learning-based AI model to detect PCa lesions on bpMRI when correlated to wholemount histopathological annotations. METHODS: Prostate final pathology specimens after RP were sectioned at 6mm increments on a mold coplanar to the axial MRI slice. All H&E stained slides were digitally scanned and annotated by genitourinary pathologists. Using a 3D U-Net-based deep neural network, we developed an AI algorithm capable of automated unit-based segmentation of prostate bpMRI and detection of its suspicious lesions algorithm performance at both lesion and patient levels was assessed using sensitivities, true detection (TD) rates and false predictions (FP) when compared to histopathological prostatic wholemount specimen. RESULTS: 25 patients were evaluated with a median age of 63 years (IQR, 58-68) and PSA level of 7.3 ng/mL (IQR, 4.9-9). Median number of wholemount prostate slides per patient was 12 (range 6-17). On histopathology, experts labeled 164 slides with a total of 1847 regions-of-interest reflecting cancerous (Gleason pattern 3/4/5, Cribriform, Signet Cells) and anatomic structures. Overall, 47 unique cancerous lesions, defined as contiguous regions across the prostate volume were identified by pathologists. 21 of these foci were detected by the AI algorithm, achieving a lesion-level sensitivity of 45%. AI had at least 1 TD in 19 out of 25 patients, resulting in a patient-level sensitivity of 76%. In total, 4 FP were made by AI, which did not correspond to any cancerous foci as defined by the pathologists. CONCLUSIONS: Our bpMRI-based AI algorithm has shown promise in identifying cancerous foci in patients with PCa. Building AI models trained with spatial annotations could enhance detection capabilities. Opportunities to improve the detection and treatment of PCa will require further development of these AI algorithms. Download PPT Source of Funding: Intramural Research Program of the NCI, NIH © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e112 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Charles Hesswani More articles by this author Enis C. Yilmaz More articles by this author Stephanie A. Harmon More articles by this author David G. Gelikman More articles by this author Christopher R. Koller More articles by this author Sahil H. Parikh More articles by this author Kyle C. Schuppe More articles by this author William S. Azar More articles by this author Daniel Nethala More articles by this author Neil Mendhiratta More articles by this author Alexander P. Kenigsberg More articles by this author Sandeep Gurram More articles by this author Baris Turkbey More articles by this author Peter A. Pinto More articles by this author Expand All Advertisement PDF downloadLoading ...