Get started
Work

Toward Placental Region Identification and Blood Vessel Classification Using Machine Learning

Year: 2022
Type: article
Abstract: Background: Purpose: To evaluate a clinical-level AI-based solution when used by pathologists for reporting and diagnosing prostate biopsies.Methods: A two-arm reader study with crossover design prospectively tested the performance of pathologists using an AI-based solution workflow compared with the use of the standard of care (pathologists using a microscope).Eight pathologists from six different sites of a large pathology laboratories network participated in the study and reported on 3,628 H&E slides from 480 consecutive prostate biopsies.Each case was reported twice, both with the microscope and with the AI-based solution randomized between pathologists.Detailed time measurements were taken for each step and compared between study arms, as was TAT.To assess the effect on the accuracy of reporting, discrepant reports were adjudicated and reviewed by a team of blinded uropathologists.Results: The study endpoints were met, specifically demonstrating that reporting with the AI-based solution leads to 37% efficiency gains and a significant decrease of 1-2 days in TAT.Importantly, case-level reporting, including tumor detection, grading, and quantitation with the AI solution was as accurate as with the microscope.The overall user experience, as reported by pathologists, was markedly better with the AI solution compared to a microscope.Conclusion: This is the first large-scale study, in which pathologists perform full primary diagnosis with the support of an integrated AI-based solution.Significant efficiency gains are observed for the pathologist in addition to the potential of improving diagnosis accuracy and harmonization.
DOI: 10.1016/j.jpi.2022.100071
Authors: Brian Vadasz, Lee A. Cooper, Jeffery A. Goldstein
Language: English
Open Access status: gold
Estimated APC paid: $1,400
OpenAlex ID: W4293216830
FWCI:
Cites:
Cited by:
Related to: 10