MP17-14 IDENTIFYING CRIBRIFORM PATTERN IN MPMRI SCANS FROM 3D RECONSTRUCTED RALP SPECIMENS USING DEEP LEARNING

Abstract

You have accessJournal of UrologyCME1 Apr 2023MP17-14 IDENTIFYING CRIBRIFORM PATTERN IN MPMRI SCANS FROM 3D RECONSTRUCTED RALP SPECIMENS USING DEEP LEARNING Ashwin Sachdeva, Dinisha Kadam, Avaneesh Meena, Nitin Singhal, Archie Parekh, Guillaume Antem, Yatin Jain, Pedro Oliveira, and Vijay Ramani Ashwin SachdevaAshwin Sachdeva More articles by this author , Dinisha KadamDinisha Kadam More articles by this author , Avaneesh MeenaAvaneesh Meena More articles by this author , Nitin SinghalNitin Singhal More articles by this author , Archie ParekhArchie Parekh More articles by this author , Guillaume AntemGuillaume Antem More articles by this author , Yatin JainYatin Jain More articles by this author , Pedro OliveiraPedro Oliveira More articles by this author , and Vijay RamaniVijay Ramani More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000003237.14AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Amongst various Gleason 4 patterns, cribriform morphology (Gl4Cri) is unique, as it seems to correlate with an aggressive clinical course, distant metastases, and disease-specific mortality. However, the true value of Gl4Cri on PCa management has recently been challenged, given that cribriform morphology is frequently intermixed with other morphologic patterns (such as glomeruloid and intra-ductal carcinoma, IDC), and biopsy-based studies suffer from sampling bias. A methodology for precise identification and quantification of Gl4Cri would help inform the role of Gl4Cri in impacting PCa management (active surveillance vs surgery vs radiotherapy) and molecular germline testing. METHODS: Using deep learning based methods, we developed an algorithm for identification and quantification of cribriform and glomeruloid patterns in whole mount images of radical prostatectomy (RP) specimens. Whole-mount H&E-stained histopathological sections of formalin-fixed RP specimens showing a mixture of different GP4 subtypes were digitised. A neural network based on U-Net++ with EfficientNet backbone was trained to identify and separate cribriform and glomeruloid glands from benign, other tumour patterns, and stromal regions. Two pathologists annotated a training set of 13 RP samples and a validation set of 42 RP samples with perfect consensus. We then applied this new algorithm to label Gl4Cri on mpMRI images using reconstructed 3D histology, allowing the extent of Gl4Cri cancer to be mapped directly onto mpMRI for development of a radiomic signature. RESULTS: The model performed well on the validation set with AUCs (Area Under Curve) of 0.87 for cribriform and 0.81 for glomeruloid pattern identification. For the two patterns, the segmentation model achieved an F1-score of 0.83 and 0.80, respectively. Using deep learning, we developed a method for automated localisation of cribriform and glomeruloid growth patterns in H&E-stained RP whole slide images, which could precisely determine the proportion of each growth pattern on corresponding mpMRI scans. CONCLUSIONS: We developed a method for automated localisation of cribriform and glomeruloid growth patterns in radical prostatectomy images and demonstrated the ability to map these patterns within pre-biopsy mpMRI scans successfully. In the future, we intend to further improve the algorithm for IDC identification and conduct multicentre trials on larger data sets for validation of this approach. Source of Funding: NHS/Aira-Matrix Project PO01 © 2023 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 209Issue Supplement 4April 2023Page: e218 Advertisement Copyright & Permissions© 2023 by American Urological Association Education and Research, Inc.MetricsAuthor Information Ashwin Sachdeva More articles by this author Dinisha Kadam More articles by this author Avaneesh Meena More articles by this author Nitin Singhal More articles by this author Archie Parekh More articles by this author Guillaume Antem More articles by this author Yatin Jain More articles by this author Pedro Oliveira More articles by this author Vijay Ramani More articles by this author Expand All Advertisement PDF downloadLoading ...