Real-World Insights in Designing SteatoStat: An End-to-End Deep Learning Pipeline for Hepatic Steatosis...

Area of Science:

• Hepatology
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Metabolic dysfunction-associated steatotic liver disease (MASLD) affects 30% of the global population.
• Current assessment of hepatic steatosis relies on subjective pathologist grading, leading to inter-observer variability.
• Objective quantification of steatosis is crucial for MASLD diagnosis and management.

Purpose of the Study:

• To develop and validate SteatoStat, a novel deep learning pipeline for standardized hepatic steatosis quantification in MASLD.
• To address the limitations of manual grading by providing an objective, reproducible measure.
• To improve the accuracy and reliability of steatosis assessment in clinical practice.

Main Methods:

• SteatoStat integrates file format standardization, rule-based cell filtering, and multiple segmentation models.
• The pipeline processes liver images to generate a continuous quantitative measure of steatosis percentage.
• Outputs are translated into standardized steatosis grades for clinical interpretation.

Main Results:

• SteatoStat achieved high performance metrics: DICE score = 0.8955, AUROC = 0.9928, F1 score = 0.8990.
• The tool demonstrated strong agreement with expert pathologists (weighted Kappa = 0.837).
• SteatoStat outperformed an existing model (weighted Kappa = 0.765), indicating superior performance.

Conclusions:

• SteatoStat provides a standardized and objective method for quantifying hepatic steatosis in MASLD.
• The pipeline shows significant potential for clinical utility in improving diagnostic accuracy.
• Future work will focus on enhancing generalizability and clinical integration through multi-institutional validation.