Quantitative Histopathological Analysis of a Diet‐Induced NASH Mouse Model Using Artificial Intelligence

Le résumé fourni par la source

Non‐alcoholic fatty liver disease (NAFLD) is one of the most common causes of chronic liver disease. It results in an accumulation of fat in the liver, and can progress to a more pathologically significant form of NAFLD known as non‐alcoholic steatohepatitis (NASH). Patients with NAFLD and NASH present on a spectrum of the disease, characterized by hepatitis (inflammation) and hepatocellular ballooning (cellular injury), which can lead to excessive fibrosis and scarring. Currently, diagnosis is confirmed by liver biopsies that are qualitatively analyzed by experienced pathologists who assign scores for each feature. However, documented inter‐pathologist variability in scoring and the semi‐quantitative nature of the scoring system itself highlight the need for new methods to differentiate NAFLD and NASH, and ensure the unbiased, consistent assessment of disease to better diagnose and stratify patients for treatment options. To support this effort we developed imageDx™: NASH, a collection of artificial intelligence (AI)‐based pathology models, to histopathologically validate and characterize NASH in a novel Diet Induced NASH BL6 Model. Livers from mice on a NASH‐inducing diet and a control diet were processed, sectioned, stained and digitally scanned to generate whole slide images. Quantification of each of the pathological features in mice was performed using imageDx™: NASH, including steatosis (macro‐ and micro‐vesicular), inflammation, hepatocellular ballooning/degeneration, fibrosis and the presence of Mallory bodies. These validated machine learning algorithms were developed with input from experienced pathologists with the goal of providing a more informative analysis of the pathology. The AI‐generated results were compared with semi‐quantitative scores provided by a board‐certified doctor of veterinary medicine. Using this diet‐induced NASH rodent model both with and without therapeutic interventions, these algorithms effectively identified small differences in tissue morphology that may not be easily discernible by visual examination. By developing highly reproducible and sensitive computational models, this study provides a foundation for a better understanding of disease mechanism and the development of effective therapeutics for NASH.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.