Introduction: The incidence rate of prostate cancer has been increasing in Asia. The number of prostate needle biopsies is also increasing for early detection and treatment. However, prostate cancer has some borderline lesions which are difficult to diagnose, that may become the burden on pathologists. Although there were some studies of artificial intelligence (AI) to support prostate cancer pathological diagnosis worldwide, there were few large-scale validations using data from Asian patients. We have developed a simple diagnostic support system that can be used immediately in clinical practice based on Asian data. Methods: A dataset of 12,230 hematoxylin-eosin stained prostate needle biopsy specimens was converted to digital images by whole slide image scanner, and annotated about cancer sites and Gleason patterns by pathologists. From these data sets, a cancer diagnosis and Gleason score discrimination system was created by deep learning. For the external validation, we obtained other 12,230 data sets from Shin-yurigaoka General Hospital, and confirmed the discrimination accuracy of cancer or non-cancer, and Gleason score based on International society of urological pathology (ISUP) grade. Results: As a result of external validation, a high detection rate of 99.85% was achieved for the sensitivity and 69.62% for specificity by a needle unit. About 1 in 1000 needles was misdiagnosed. Most of these were cancerous ducts smaller than 1 mm or small ducts in the inflammatory background. Aggregating the results based on a patient, the cancer detection rate was 100%. On the other hand, the discrimination rate for ISUP grade was following, ISUP1:58.5%, ISUP2:31.1%, ISUP3:33.6%, ISUP4:25.9%, and ISUP5:81.9%. Conclusions: The results of the external validation showed that the discrimination of cancerous areas could be diagnosed with high sensitivity, which may prevent the misdiagnosis of cancerous areas and contribute to the reduction of the pathologist's working time. However, the Gleason score was not sufficiently discriminated due to inter-observer variability. Therefore, pathologist's visual inspection was considered essential for Gleason score evaluation. SOURCE OF Funding: None
