Microscopic examination of prostate cancer has failed to reveal a reproducible association between molecular and morphologic features. However, deep learning algorithms trained on hematoxylin and eosin (H&E)-stained whole slide images (WSI) may outperform the human eye and help to screen for clinically-relevant genomic alterations. We created deep learning algorithms to identify prostate tumors with underlying ERG fusions or PTEN deletions using four stages: A) automated tumor identification; B) feature representation learning; C) classification; and D) explainability map generation. A novel transformer-based hierarchical architecture was trained on a single representative WSI of the dominant tumour nodule from a radical prostatectomy (RP) cohort with known ERG/PTEN status (n=224 and n=205, respectively). Two distinct vision transformer (ViT)-based networks were utilized for feature extraction and a distinct transformer-based model was used for classification. ERG algorithm performance was validated across three RP cohorts, including 64 WSI held out from the pre-training cohort (area under receiver operator characteristic curve or AUC: 0.91), and 248 and 375 WSI from two independent RP cohorts (AUC: 0.86 and 0.89). In addition, we tested ERG algorithm performance in two needle biopsy cohorts comprised of 179 and 148 WSI (AUC: 0.78 and 0.80). Focusing on cases with homogeneous (clonal) PTEN status, PTEN algorithm performance was assessed using 50 WSI held out from the pre-training cohort (AUC: 0.81) as well as 201 and 337 WSI from two independent RP cohorts (AUC: 0.72 and 0.80) and 151 WSI from a needle biopsy cohort (AUC: 0.75). For explainability, the PTEN algorithm was also applied to 19 WSI with heterogeneous (subclonal) PTEN loss, where the percent tumor area with predicted PTEN loss correlated with that based on immunohistochemistry (r=0.58, p=0.0097). These deep learning algorithms to predict ERG/PTEN status provide proof-of-principle that H&E images can be used to screen for underlying genomic alterations in prostate cancer.
Modern pathology : an official journal of the United States and Canadian Academy of Pathology, Inc. 2023 Jun 10 [Epub ahead of print]
Eric Erak, Lia DePaula Oliveira, Adrianna A Mendes, Oluwademilade Dairo, Onur Ertunc, Ibrahim Kulac, Javier A Baena-Del Valle, Tracy Jones, Jessica L Hicks, Stephanie Glavaris, Gunes Guner, Igor Damasceno Vidal, Mark Markowski, Claire de la Calle, Bruce J Trock, Avaneesh Meena, Uttara Joshi, Chaith Kondragunta, Saikiran Bonthu, Nitin Singhal, Angelo M De Marzo, Tamara L Lotan
Department of Pathology, Johns Hopkins University School of Medicine, United States., Department of Pathology, Suleyman Demirel University, Turkey., Koç University School of Medicine, Turkey., Fundacion Santa Fe de Bogota University Hospital, Columbia., Hacettepe University, Turkey., Department of Pathology, University of Alabama School of Medicine, United States., Department of Oncology, Johns Hopkins University School of Medicine, United States., Department of Urology, Johns Hopkins University School of Medicine, United States., AIRA Matrix Private Limited, India., Department of Pathology, Johns Hopkins University School of Medicine, United States;; Department of Oncology, Johns Hopkins University School of Medicine, United States;; Department of Urology, Johns Hopkins University School of Medicine, United States., Department of Pathology, Johns Hopkins University School of Medicine, United States;; Department of Oncology, Johns Hopkins University School of Medicine, United States;; Department of Urology, Johns Hopkins University School of Medicine, United States;. Electronic address: .