AI-Based Models for Risk Prediction in MASLD: A Systematic Review

Area of Science:

• Medical Informatics
• Hepatology
• Artificial Intelligence

Background:

• Metabolic dysfunction-associated steatotic liver disease (MASLD) is a growing global health issue.
• Early identification of patients at risk for significant disease progression (e.g., advanced fibrosis, MASH) is critical for effective management.
• A gap exists in understanding the efficacy of AI in MASLD risk prediction and patient stratification.

Purpose of the Study:

• To systematically review and evaluate the performance of AI-based models in predicting MASLD risk.
• To assess the ability of AI models to stratify patients based on disease severity, including fibrosis and MASH.
• To identify key predictors and common methodologies used in AI for MASLD risk assessment.

Main Methods:

• A comprehensive systematic literature search was conducted following PRISMA guidelines across major databases.
• Studies were included based on predefined criteria for AI-based MASLD risk prediction.
• Data extraction and quality assessment (QUADAS-2) were performed, with the study registered in PROSPERO.

Main Results:

• 26 studies (2014-2025) from diverse geographical regions were included, with predominantly low risk of bias.
• AI models demonstrated strong predictive performance: AUROCs for MASH ranged from ~0.76-0.95, and for fibrosis (≥F2-F4) from ~0.72-0.94.
• Common predictors included age, BMI, liver enzymes, and platelets; multimodal data (clinical, imaging, elastography) often improved discrimination.

Conclusions:

• AI-based models show significant promise for predicting MASLD risk and stratifying patients.
• The predictive capabilities of these AI models are robust, offering potential for improved clinical decision-making.
• Enhancing data diversity and model interpretability are key future steps for successful clinical implementation of AI in MASLD.