- AIM-MASH AI Assist* is a deep learning-based digital pathology tool that predicts MASH clinical research network (CRN)-aligned grades and stages from input whole slide images of MASH biopsies based on pixel-level quantification of steatosis, hepatocellular ballooning, lobular inflammation, and fibrosis [1,2]. In prospective clinical trials, this tool is intended to assist pathologists with scoring for enrollment and endpoint assessment.
- AIM-MASH demonstrates superior accuracy for scoring hepatocellular ballooning and lobular inflammation compared to a gold-standard pathologist consensus, along with comparable accuracy for scoring steatosis and fibrosis [2].
- The United States Food and Drug Administration (FDA) recently qualified AIM-MASH AI Assist to aid pathologist evaluation of MASH in clinical trials [3]. The FDA qualification stipulated that AIM-MASH AI Assist be used in a gold-standard consensus process, with consensus calls necessary for cases with inter-pathologist disagreement (Figure 1).
- Here, we evaluated this workflow to inform the design of MASH clinical trials incorporating AIM-MASH AI Assist, with the following objectives:
Establish the effect of reader harmonization on inter-reader agreement.
Evaluate inter-reader agreement using manual and AI-assisted review.
Assess the inter-panel agreement of 2-pathologist and 3-pathologist panels.
Compare the percent modal agreement between 2-pathologist and 3-pathologist panels.
Conference
Paris Mash Meeting
View Poster