Leveraging Machine Learning for Equitable Screening of Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD)

Overview

Metabolic Dysfunction-Associated Steatotic Liver Disease (MASLD) affects 33.7% of American adults, yet early detection remains challenging because diagnostic methods like FibroScan® are expensive and inaccessible. FibroScan® provides objective estimates of liver steatosis (fatty liver) and fibrosis (scarring). Early detection is critical as initial stages of MASLD are reversible through lifestyle changes, making accessible screening tools essential for public health. This project hypothesized that machine learning could accurately predict MASLD using accessible demographic and laboratory data, enabling cost-effective population screening. Data from 11,789 subjects (NHANES 2017–2023) with FibroScan® measurements were analyzed to develop two XGBoost binary classifiers: Model-1.0 (demographics-only) and Model-2.0 (demographics + metabolic markers). Model-1.0 utilized eight demographic features including BMI, waist-circumference, age, and ethnicity, while Model-2.0 added three liver enzyme biomarkers routinely measured in annual blood draws: ALT, AST, and ALP. Local validation was conducted using FibroScan® on a cohort of 21 patients at Gastroenterology & Liver Institute. Model-1.0 achieved 83.4% accuracy for steatosis prediction with robust performance across BMI categories and ethnic groups (80%+ accuracy), demonstrating the model's ability to learn population-specific patterns. Model-2.0 achieved 84.5% accuracy for steatosis and 81.1% accuracy for fibrosis, with leading predictors being waist-circumference, ethnicity, BMI, and ALT. The integration of accessible liver enzyme biomarkers significantly boosted fibrosis prediction (+25%), from unreliable to clinically applicable, approaching the screening benchmarks of commercial tests like FibroSure. Local validation confirmed real-world applicability with 76.2% accuracy. Our dual-model approach enables cost-effective MASLD screening in routine checkups, potentially saving patients $200,000+ per 1,000 individuals screened by reducing reliance on expensive diagnostic tests. Model-1.0 suits initial population-level screening using only physical measurements, while Model-2.0 provides comprehensive assessment of both steatosis and fibrosis when basic metabolic panels are available.

Competition history

• CSEF 2026 Medicine & Physiology (Track 2) (Senior Division) · Entry S-21-25

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Biomedical and Health Sciences projects
• CSEF Biomedical and Health Sciences projects
• Every CSEF 2026 project