B57: Automated Quantification of Whole Slide Steatosis Using a Custom LiverQuant Pipeline

Presenter(s)

Evan Mayse

Abstract or Description

The histological gold standard for assessing hepatic steatosis relies on visual estimation by a trained pathologist, a process that is time-intensive and subject to observer variability. Automated image analysis offers opportunities to improve scalability, reproducibility, and throughput in liver disease research. This study evaluates a custom Python-based pipeline developed to implement and extend the LiverQuant framework in a murine model of intestinal failure-associated liver disease. Hematoxylin and eosin-stained whole-slide images were obtained from a longitudinal cohort of male and female mice undergoing sham operation or 75% small bowel resection, harvested at 2, 5, 10, and 26 weeks postoperatively, and scanned at 20x magnification using an Olympus VS120 whole-slide scanner. A total of 110 slides were analyzed. A custom pipeline was developed to convert whole-slide images, generate a tissue mask via color thresholding, and quantify steatosis as the percentage of lipid area relative to total tissue area. Automated measurements were compared with pathologist-derived estimates of steatosis using Pearson correlation analysis. Automated quantification demonstrated moderate-to-strong correlation with pathologist scoring across all time points (2wk: n=35, r=0.85; 5wk: n=29, r=0.78; 10wk: n=19, r=0.70; 26wk: n=27, r=0.67). Batch processing of 35 slides required approximately 39 minutes on a workstation and approximately one hour on a consumer-grade laptop. Pipeline performance was consistent across male and female cohorts. The custom LiverQuant-based pipeline provides an efficient and reproducible approach to whole-slide steatosis quantification with performance comparable to expert pathologist assessment, supporting its use as a scalable tool for experimental liver pathology research.