PD50-06 PERFORMANCE OF A PROSTATE CANCER DETECTION SYSTEM ON IHC-REQUESTED CASES
Ramin Nateghi, Madeline Saft, Eric V. Li, Sai Kaushik SR. Kumar, Clayton Neill, Hiten Dilip Patel, Edward Matthew Schaeffer, Ximing J. Yang, Ashley Evan Ross, Lee Cooper · The Journal of Urology · 2024
You have accessJournal of UrologyProstate Cancer: Detection & Screening V (PD50)1 May 2024PD50-06 PERFORMANCE OF A PROSTATE CANCER DETECTION SYSTEM ON IHC-REQUESTED CASES Ramin Nateghi, Madeline Saft, Eric V. Li, Sai Kumar, Clayton Neill, Hiten Patel, Edward Schaeffer, Ximing J. Yang, Ashley E. Ross, and Lee A. D. Cooper Ramin NateghiRamin Nateghi , Madeline SaftMadeline Saft , Eric V. LiEric V. Li , Sai KumarSai Kumar , Clayton NeillClayton Neill , Hiten PatelHiten Patel , Edward SchaefferEdward Schaeffer , Ximing J. YangXiming J. Yang , Ashley E. RossAshley E. Ross , and Lee A. D. CooperLee A. D. Cooper View All Author Informationhttps://doi.org/10.1097/01.JU.0001008620.35181.96.06AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Triple immunostain (AMACR, p63 and high molecular keratin) is utilized by pathologists when the diagnosis of prostate cancer is in question. Augmented intelligence utilized on digital pathology can help confirm the histological diagnosis of prostate cancer and potentially reduce the need for additional immunohistochemical analysis such as with the triple stain (PIN4) cocktail. Here we evaluated the accuracy of an institutionally developed AI algorithm to detect prostate cancer among H+E stained biopsy slides from cases which were that PIN4 staining was utilized. METHODS: Two datasets were evaluated: a "Prostate Cancer Cohort'' containing whole slide images of H+E stained slides from 4703 tissue blocks. These blocks contain biopsy cores taken from 509 patients diagnosed with prostate adenocarcinoma, and a "PIN4 cohort" including H+E stained slides from 182 tissue blocks with supplementary triple stained PIN4 testing (Table 1). All cases were reviewed histologically by an expert genitourinary pathologist.We developed a weakly supervised deep learning method that employs an attention-based multiple-instance learning model to automatically detect prostate cancer based on slide-level annotations. The prostate cancer cohort was split into three sets: training, validation, and testing, with an 80/10/10 ratio. The developed model was then tested on H+E slides corresponding to biopsies that underwent PIN4 testing. RESULTS: Our model demonstrated an Area Under the Curve (AUC) of 97.92% for detection of prostate cancer within the prostate cancer cohort. Among diagnostically challenging triple stained cases the locked model correctly predicted prostate cancer with an AUC of 80.7% (Figure 1). At a confidence level of 90% and a false positive rate of less than 1%, 37% of triple staining have been omitted by our AI model. CONCLUSIONS: AI algorithms can accurately detect prostate cancer from digitized H+E stained slides. Use of our AI algorithm has the potential to reduce utility of triple stain which can improve workflow for pathologists and limit cost for patients and the health care system. Download PPT Source of Funding: Polsky Urologic Cancer Institute Award © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e1059 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Ramin Nateghi More articles by this author Madeline Saft More articles by this author Eric V. Li More articles by this author Sai Kumar More articles by this author Clayton Neill More articles by this author Hiten Patel More articles by this author Edward Schaeffer More articles by this author Ximing J. Yang More articles by this author Ashley E. Ross More articles by this author Lee A. D. Cooper More articles by this author Expand All Advertisement PDF downloadLoading ...