Weakly supervised and transfer learning for adenocarcinoma classification in transurethral resection of the prostate whole slide images

Abstract

AbstractThe transurethral resection of the prostate (TUR-P) is generally considered an option for benign prostatic diseases especially nodular hyperplasia patients who have moderate to severe urinary problems that have not responded to medication. Importantly, incidental prostate cancer are diagnosed at the time of TUR-P for benign prostatic disease. Since diagnosing a large number of cases containing TUR-P specimens which are characterized by a very large volume of tissue fragments by pathologists using a conventional microscope is time-consuming and limited in terms of human resources. Thus, it is necessary to develop new techniques which can rapidly and accurately screen large numbers of TUR-P specimens. Computational pathology applications which can assist pathologists in detecting prostate adenocarcinoma from TUR-P whole slide images (WSIs) would be of great benefit for routine histopathological workflow. In this study, we trained deep learning models to classify TUR-P WSIs into prostate adenocarcinoma and benign (non-neoplastic) lesions using transfer and weakly supervised learning. We evaluated the models on TUR-P, needle biopsy, and The Cancer Genome Atlas (TCGA) public dataset test sets, achieving an ROC-AUC up to 0.984 in TUR-P test sets for adenocarcinoma. The results demonstrate the high promising potential of deployment in a practical TUR-P histopathological diagnostic workflow system.