Validation, implementation, and impact of an AI model in routine practice for pathologic diagnosis of prostate cancer in an academic medical center.

BACKGROUND

Pathological evaluation of prostate needle biopsies is labor-intensive and often requires ancillary immunohistochemistry (IHC), increasing cost and diagnostic turnaround time (TAT). Artificial intelligence (AI)-based decision-support tools may improve efficiency, but clinical deployment requires institutional validation and assessment of real-world impact.

METHODS

Within a fully digital pathology practice, we performed institutional validation of an AI-assisted prostate biopsy decision-support tool using routine clinical cases, with pathologist-rendered diagnoses as ground truth. Following validation, the tool was implemented into the routine sign-out. A retrospective pre-post analysis compared prostate biopsy cases signed out during 3-month periods before and after implementation, excluding a transition month. Diagnostic TAT was defined as the interval from whole-slide image scan completion to final sign-out. IHC utilization was recorded. Weighted median TATs and IHC use were compared using standard statistical methods.

RESULTS

The validation cohort met all predefined acceptance criteria, demonstrating high AI performance (sensitivity 91-100%, specificity 99%, positive-predictive value 98%, negative-predictive value 96%, area under the curve 0.97). Following clinical implementation, diagnostic TAT decreased by 30% and IHC utilization decreased by 38%.

CONCLUSIONS

Institutional validation and clinical implementation of an AI-assisted prostate biopsy decision-support tool were associated with significant reductions in diagnostic TAT and IHC utilization. When deployed as an adjunct within a digital workflow, AI assistance may enhance efficiency while preserving pathologist responsibility for final diagnosis.