Artificial intelligence scoring of liver biopsies in a phase ii trial of semaglutide in non-alcoholic steatohepatitis

Background and Aims: Artificial intelligence-powered digital pathology offers the potential to quantify histologic findings in a reproducible way. This analysis compares the evaluation of histological features of non-alcoholic steatohepatitis (NASH) between pathologists and a machine learning (ML) pathology model. Approach and Results: This post hoc analysis included data from a subset of patients (N=251) with biopsy-confirmed NASH and fibrosis stage F1–F3 from a 72-week randomised placebo-controlled trial of once-daily subcutaneous semaglutide 0.1, 0.2, or 0.4 mg (NCT02970942). Biopsies at baseline and week 72 were read by two pathologists. Digitised biopsy slides were evaluated by PathAI’s NASH ML models to quantify changes in fibrosis, steatosis, inflammation, and hepatocyte ballooning using categorical assessments and continuous scores. Pathologist and ML-derived categorical assessments detected a significantly greater percentage of patients achieving the primary endpoint of NASH resolution without worsening of fibrosis with semaglutide 0.4 mg vs placebo (pathologist 58.5% vs 22.0%, p <0.0001; ML 36.9% vs 11.9%; p =0.0015). Both methods detected a higher but non-significant percentage of patients on semaglutide 0.4 mg vs placebo achieving the secondary endpoint of liver fibrosis improvement without NASH worsening. ML continuous scores detected significant treatment-induced responses in histologic features, including a quantitative reduction in fibrosis with semaglutide 0.4 mg vs placebo ( p =0.0099) that could not be detected using pathologist or ML categorical assessment. Conclusions: ML categorical assessments reproduced pathologists’ results of histological improvement with semaglutide for steatosis and disease activity. ML-based continuous scores demonstrated an antifibrotic effect not measured by conventional histopathology.