The potential use of artificial intelligence in reviewing atypical small acinar proliferation in prostate core biopsy.

Abstract

Prostate cancer is the most common diagnosed cancer and the second leading cause of cancer death among men in the United States. The gold standard for the diagnosis of prostate cancer is the examination of tissue from a systemic 12- or 14-core biopsy procedure by an anatomic pathologist. Atypical small acinar proliferation in prostate core biopsy is an indefinite diagnosis, and may cause difficulties in clinical management. To investigate the potential use of artificial intelligence in reviewing atypical small acinar proliferation, Paige Prostate, an FDA-approved artificial intelligence-based digital diagnostic system, was used to review the digital images of 107 cores with an initial diagnosis of atypical small acinar proliferation. Two pathologists blindly reclassified the cores with a new diagnosis (either benign or malignant) for 85% of the cases (pathologist #1) and 88% (pathologist #2), respectively. Paige Prostate resulted in consistent diagnosis in 85% of the cores ("suspicious" or "not suspicious"). The consistency rates were 77% between the two pathologists, 66% and 75% between Paige Prostate and each pathologist. There was statistically significant agreement among the three consistency rates. Although some degree of interobserver discrepancy still exists, our findings indicate that artificial intelligence was compatible with the pathologist review, and it can be a unique diagnostic adjunct for pathologists.