Interpretable Machine Learning Models for Early Detection of Metabolic Dysfunction-Associated Steatotic Liver Disease...

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Liver Disease Diagnostics

Background:

• Metabolic dysfunction-associated steatotic liver disease (MASLD) is a growing health concern.
• Early detection and risk stratification are crucial for managing MASLD.
• Non-invasive methods are needed to supplement traditional diagnostic tools.

Purpose of the Study:

• To develop and evaluate interpretable machine learning models for early MASLD detection.
• To utilize routine clinical and laboratory data for MASLD risk stratification.
• To assess the feasibility of using MRI-PDFF-calibrated models in clinical settings.

Main Methods:

• Retrospective analysis of 152 patients' clinical and laboratory data.
• Development and comparison of ten interpretable machine learning models.
• Validation using magnetic resonance imaging proton density fat fraction (MRI-PDFF) and Shapley Additive Explanations (SHAP).

Main Results:

• Logistic Regression showed high discriminative performance (AUC = 0.873).
• Random Forest achieved the best overall classification (accuracy = 0.816, F1-score = 0.807).
• Key predictors identified: Hepatic Steatosis Index (HSI), body fat, age, and triglycerides (TG).

Conclusions:

• Interpretable machine learning models using routine data are feasible for MASLD risk stratification.
• SHAP analysis identified significant metabolic and anthropometric risk factors.
• This approach can help prioritize patients for imaging and further evaluation.