Ström P, Kartasalo K, Olsson H, et al. - Researchers intended to construct an artificial intelligence (AI) system that can assist in prostate cancer detection, localisation, and Gleason grading, with clinically acceptable accuracy, so that overtreatment and undertreatment of prostate cancer resulting from high intra-observer and inter-observer variability in grading may be prevented. For this purpose, they used needle core biopsies from 976 randomly elected participants aged 50–69 in the Swedish prospective and population-based STHLM3 diagnostic study, and also used samples from 93 men outside the study. On digitalizing the slides from needle core biopsies, the resulting images were utilized to train deep neural networks for the evaluation of prostate biopsies. For an independent test dataset including 1,631 biopsies from 246 men from STHLM3 and an external validation dataset comprising 330 biopsies from 73 men, the presence, extent, and Gleason grade of malignant tissue were predicted in order to assess the networks. For differentiating benign from malignant biopsy cores, an area under the receiver operating characteristics curve of 0·997 and 0·986 was achieved with AI on the independent test dataset and on the external validation dataset, respectively. It was concluded that detection and grading of cancer in prostate needle biopsy samples is possible via a trained AI system, with a ranking comparable to that of international experts in prostate pathology. By decreasing the evaluation of benign biopsies and by automating the task of measuring cancer length in positive biopsy cores, this system could decrease pathology workload when applied clinically. Besides, a second opinion and standardising grading might be achieved with an AI system with expert-level grading performance. It might also afford pathology expertise in parts of the world where it does not exist.