Formative Usability Testing of Artificial Intelligence in Pathology: The Challenge of Assessing Acceptability

Abstract

Digital Pathology has provided a platform to use Artificial Intelligence (AI) to assist pathologists with diagnosis and reporting. An AI tool is being developed that analyzes digital Hematoxylin and Eosin (stained tissue) images associated with a skin cancer case and pre-populates a report with required parameters. The aim of this AI pathology assistant is to save pathologist time and increase reporting efficiency. This study assessed ease of use and acceptability of a first iteration of the AI tool. Twelve pathologists were recruited across seven UK hospitals and participated in a think-aloud evaluation, completing a pathology report using the novel tool, after which they participated in a brief interview. The think-aloud identified several issues that can inform tool development to improve ease of use. AI performance (inaccuracy populating report items) constrained assessment of tool acceptability and added tasks to the reporting process. This finding emphasizes the importance of AI accuracy (1) for assessing if and how such tools can be integrated into clinician's workflow to increase efficiency, and (2) for cultivating clinician trust in tool performance to support adoption in practice.

Ähnliche Arbeiten