A Comparison of AI-based NASH CRN scoring to manual methods in Phase 2 clinical trials


Analyze the performance of AI-based Measurement of NASH Histology (AIM-NASH) endpoint analysis from two recent Phase 2 NASH clinical trials


Manual scoring of NASH histopathology is subject to observer variability.

  • PathAI developed its AIM-NASH algorithm using over 5,900 biopsies from 6 completed NASH clinical trials.
  • AIM-NASH was used in the retrospective analysis of two recent Phase 2 trials (n=411 liver biopsy samples) for endpoint assessment. Results were compared to those obtained via manual scoring.


A significantly greater number of primary endpoint responders is detected in treated vs. placebo arms by AIM- ASH but not manual scoring in Ph2 study of Pegbelfermin (PGBF)

Primary endpoint responders were patients with greater to or equal to 1 stage NASH CRN fibrosis improvement with no worsening of fibrosis at week 24.

AIM-NASH detected a dose-dependent treatment response in a Phase 2 NASH clinical study, in addition to a statistically significant difference in the proportion of patients achieving endpoints when treated with 0.4mg

*Semaglutide relative to untreated patients


In an endpoint analysis, AIM-NASH enabled statistically significant results and dose dependencies, which were not observed using traditional manual histology endpoint analysis. AIM-NASH may support future prospective NASH clinical trials to enable accuracy and confidence in endpoint assessment and measurement of treatment effect.

Study Participants and Design Summary

  1. Shevell et al. Comparison of manual vs machine learning approaches to liver biopsy scoring for NASH and fibrosis: a post hoc analysis of the FALCON 1 study. AASLD poster 2021

    • FALCON 1 was a phase 2b, randomized, multicenter, placebo-controlled study assessing the efficacy and safety of pegbelfermin (PGBF)

    • Eligible adults were 18-75 years of age with a liver biopsy tissue specimen collected within 6 months prior to or during screening that was consistent with NASH with a score of greater to or equal to 1 for each NAS component and stage 3 liver fibrosis according to the NASH CRN classification

    • Primary endpoint responders were patients with greater to or equal to 1 stage NASH CRN fibrosis improvement without NASH worsening or NASH improvement with no worsening of fibrosis at week 24. Conchran-Armitage test for trend was used to comprare PGBF vs placebo.
  1. Harrison et al. Artificial intelligence-powered digital pathology model supports that fibrosis is reduced by semaglutide in patients with NASH. AASLD poster 2021

    • Using liver biopsy samples from a phase 2 trial that compared the effect of three different doses of the glucagon-like peptide-1 receptor agonist semaglutide (0.1, 0.2 and 0.4 mg once daily) with placebo in patients with NASH

    • All biopsies at baseline (N=320) and week 72 were each manually assessed by two independent pathologists who were blinded to the patient, treatment and each other’s assessment. During trial conduct, digitalization of liver biopsies was initiated and 251 subjects had their baseline biopsy digitalized.

 AIM-NASH is currently under review as a drug development tool by FDA and EMA

Contact us to learn more about
opportunities to deploy AIM-NASH in
clinical trials