As published in Sage Journals, researchers have developed an accessible, machine learning-driven histopathology workflow that enables quantitative evaluation of metabolic dysfunction-associated steatohepatitis (MASH) in laboratory mice without requiring coding expertise or proprietary software.
Histologic image analysis is becoming increasingly important in biomedical and comparative pathology, but adoption has often been slowed by technical barriers. To address this challenge, investigators created a liver pathology analysis pipeline using QuPath, a freely available image analysis platform with a graphical user interface, making advanced digital pathology techniques more accessible to veterinary pathologists and researchers.
Building a Machine Learning Classifier
The team trained a pixel-based classifier using annotations from seven digitized hematoxylin and eosin (H&E)-stained liver sections obtained from a diet-induced mouse model of MASH. The classifier was designed to distinguish among several key tissue features, including:
- Normal hepatocytes
- Macrovesicular steatosis
- Microvesicular steatosis
- Blood vessels
- Inflammatory areas
To further quantify disease-related changes, the researchers applied image-segmentation and thresholding techniques to measure lipid droplet accumulation within steatotic regions. Additional threshold-based analyses were performed on picrosirius red-stained sections to assess liver fibrosis.
Strong Agreement With Conventional Pathology Scoring
The quantitative measurements generated by the image-analysis workflow were compared with traditional histopathologic evaluations based on the nonalcoholic steatohepatitis activity score (NAS).
Mice fed the MASH-inducing diet exhibited substantially greater areas of macrosteatosis than control animals maintained on a standard diet. Moreover, digital measurements of steatosis, inflammatory infiltration, and fibrosis showed strong associations with increasing histopathology scores, supporting the validity of the computational approach.
Advantages Over Ordinal Scoring Systems
Unlike conventional grading methods, which categorize lesions into discrete severity levels, the QuPath-based workflow produces continuous numerical data. This allows researchers to perform more robust statistical analyses and detect subtle disease-related differences that may be difficult to capture with ordinal scoring alone.
The authors emphasize that the method is scalable, reproducible, and user-friendly. Because it relies on open-source software and does not require programming skills, the approach could broaden access to quantitative digital pathology tools in both veterinary and biomedical research settings.
Implications for MASH Research
As interest in MASH continues to grow, particularly in preclinical therapeutic development, reliable and objective measures of liver injury are increasingly important. This study demonstrates that machine learning-assisted image analysis can provide detailed quantitative assessments of steatosis, inflammation, and fibrosis while remaining accessible to investigators without specialized computational training.
The workflow offers a practical framework for generating high-resolution quantitative pathology data and may help improve the consistency and interpretability of findings in future MASH studies.
