An artificial intelligence solution for evaluation of prostate needle core biopsy specimens: impact in a non-specialist setting

The diagnosis of prostate cancer rests on the histopathological evaluation of prostate needle core biopsy specimens (NCBS) supplemented by immunohistochemistry as required. Because of the well-known interobserver variability in diagnosis and grading of prostate cancer, we have studied the capability of a commercially available artificial intelligence (AI) solution to aid in the diagnosis of PCa in a non-expert setting. For this, 2828 H&E stained NCBS slides from 249 cases were digitised to produce whole slide images (WSI) and subjected to second-read analysis using the Ibex prostate AI solution. Ground truth was established by a combination of primary pathologists' diagnoses supplemented by immunohistochemistry and external expert revision. For cancer detection, the AI solution achieved a false negative rate of 0.2%, significantly lower than that of the reporting pathologists (1.5%). False positive rates were similar for reporting pathologists (2.3%) and prostate AI solution (3.1%). A Gleason pattern 4 alert was raised by the AI solution in 2 of 29 cases (7%) of cases originally diagnosed as Gleason score 3 + 3, both of which were upheld after expert review. Because of the therapeutic consequences, the triggering of a "higher than 3 + 3 Gleason score" alert represents a useful safety feature. In conclusion, our findings support the growing evidence that by enhancing diagnostic accuracy AI-based diagnostic solutions can play a pivotal role in alleviating the increasing workload faced by pathologists due to the rising global incidence of prostate cancer.