AI-Powered Image Analysis in Pathology: Integrating Custom Tools for Faster Diagnosis

Abstract

Diagnostic pathology faces rising slide volumes, workforce shortages and inter-observer variability, while whole-slide imaging has made computational analysis feasible at scale. This study developed and evaluated a custom AI image-analysis toolchain — combining tissue detection, a deep-learning region-of-interest classifier, and a pathologist-facing overlay interface — and measured its effect on diagnostic speed and accuracy in a computer-assisted (AI-assisted) reading workflow. Across 1,200 whole-slide images spanning breast lymph-node and prostate biopsy cohorts, the standalone classifier achieved an area under the receiver-operating-characteristic curve (AUROC) of 0.974. In a reader study of eight pathologists, AI-assisted reading reduced mean time per slide from 92.4 to 41.7 seconds (−54.9%), raised sensitivity from 91.3% to 97.6%, and cut major discordances by 62%. A pre-screening configuration allowed 68.4% of clearly negative slides to be safely deprioritised at 100% sensitivity. Integrated custom AI tooling can meaningfully accelerate pathology diagnosis without sacrificing accuracy.