Pixel classifier-based image analysis of metabolic dysfunction-associated steatohepatitis in mice

Image analysis in histologic assessments has increasingly gained traction in comparative pathology and biomedical research. However, application has been limited by a knowledge gap, particularly for veterinary pathologists without a strong background in coding-intensive image analysis. This study describes a quantitative image analysis pipeline for liver pathology in a special diet-induced metabolic dysfunction-associated steatohepatitis (MASH) mouse model, using the existing open-source, graphic user interface-based program QuPath. A machine learning-based pixel classifier was trained on annotations made in 7 digitized hematoxylin and eosin (H&E)-stained slides to detect normal hepatocytes, macrosteatosis, microsteatosis, vessels, and inflammation. Macrosteatosis annotations were then segmented via pixel thresholding to quantify lipid droplets. Post-processing pixel thresholding was applied to quantify fibrosis in picrosirius red-stained slides and lipid droplets in H&E-stained slides. Results were benchmarked against traditional manual grading by the nonalcoholic steatohepatitis activity score (NAS). Our results revealed a significantly larger macrosteatosis area percentage in MASH diet mice than in regular diet controls. The percentage areas for steatosis, inflammation, and fibrosis strongly correlated with increasing histopathologic scores. Unlike traditional ordinal severity scoring like NAS, this approach generates continuous, quantitative data suitable for parametric statistical analysis. This method is scalable and accessible, requires no coding expertise or commercial software, and offers a blueprint for generating truly quantitative histopathology data to veterinary pathologists and others engaged in MASH research.