Pd45-02 Improving Training Outcomes in AI-Enabled Multispectral Imaging for Cancer Detection and Localization in Robotic-Assisted Radical Prostatectomy

You have accessJournal of UrologyProstate Cancer: Staging II (PD45)1 May 2024PD45-02 IMPROVING TRAINING OUTCOMES IN AI-ENABLED MULTISPECTRAL IMAGING FOR CANCER DETECTION AND LOCALIZATION IN ROBOTIC-ASSISTED RADICAL PROSTATECTOMY Ahmed Gamal, Marcio Covas Moschovas, Shady Saikali, Abdel Rahman Jaber, Travis Rogers, Aminollah Khormali, O'Connor Timothy, Morgan McNamara, David Fournier, and Vipul Patel Ahmed GamalAhmed Gamal , Marcio Covas MoschovasMarcio Covas Moschovas , Shady SaikaliShady Saikali , Abdel Rahman JaberAbdel Rahman Jaber , Travis RogersTravis Rogers , Aminollah KhormaliAminollah Khormali , O'Connor TimothyO'Connor Timothy , Morgan McNamaraMorgan McNamara , David FournierDavid Fournier , and Vipul PatelVipul Patel View All Author Informationhttps://doi.org/10.1097/01.JU.0001008792.09108.b4.02AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: MarginASSURE Imaging System developed by CytoVeris uses a Multi-Spectral Tissue Auto-Fluorescence Imaging (MS-TAFI) concept and employs machine learning algorithms for tumor detection on freshly excised tissue real-time. This platform can capture and analyze the unique "optical fingerprint" of tissue based on the intrinsic biomolecular and morphological tissue characteristics, without requiring the use of dyes or imaging agents. The system's AI-generated output displays the presence or absence of cancer on imaged specimens. The objective is to train the MS-TAFI to acquire the optical fingerprint of prostate tumor tissue which can be used to visualize cancer margins. METHODS: This is a prospective study performed on prostatectomy specimens from Sep. 2023 to Oct. 2023. We obtained fresh tissue blocks from 13 patients, for each patient a minimum of two blocks were imaged using MarginASSURE: one containing tumor tissue, and the other predominately normal tissue. These 13 patients, comprising 28 tissue blocks, were used to develop a machine learning pipeline by comparing the obtained specimens with postoperative histopathology assessments (tumor annotations) to curate ground-truth labeled data. RESULTS: The obtained results from the AI model produced a 83.7% accuracy, 91.1.0% AUC, 84.0% specificity, and 83.5% sensitivity based on a Leave One Patient Out training and testing process. The developed AI algorithm offers tumor localization with high concordance with ground-truth histopathology assessment. CONCLUSIONS: Initial results showed that the system has the potential to differentiate between cancerous and benign tissue in the prostate cancer model, which may enhance the precision of tumor resection. We described acceptable sensitivity and specificity rates of cancer detection, next steps would include building a more comprehensive database for algorithm training/testing. Source of Funding: No Fund © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e967 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Ahmed Gamal More articles by this author Marcio Covas Moschovas More articles by this author Shady Saikali More articles by this author Abdel Rahman Jaber More articles by this author Travis Rogers More articles by this author Aminollah Khormali More articles by this author O'Connor Timothy More articles by this author Morgan McNamara More articles by this author David Fournier More articles by this author Vipul Patel More articles by this author Expand All Advertisement PDF downloadLoading ...