A stacking ensemble machine learning model built on routinely available clinical data demonstrated good discrimination for identifying ultrasound-defined metabolic dysfunction-associated steatotic liver disease in children with overweight or obesity, according to a retrospective multicenter study published in BMC Medical Informatics and Decision Making.
Metabolic dysfunction-associated steatotic liver disease increasingly accompanies childhood obesity, establishing an ongoing clinical need for accessible screening approaches that can be integrated into routine pediatric care.
Researchers conducted a retrospective multicenter investigation utilizing outpatient records from two hospital centers to evaluate overweight and obese children aged 5 to 16 years. The model development cohort included 7,454 pediatric patients from Nanjing, comprising 3,475 disease events and 3,979 non-events, while an independent external validation set included 353 children from Suzhou, consisting of 129 events and 224 non-events.
The study determined steatotic liver disease status through abdominal ultrasound evaluations. Across the evaluated clinical records, investigators gathered 46 routine clinical variables spanning anthropometric measurements, standard blood tests, and biochemical indicators to construct a stacking ensemble framework combining ten machine learning algorithms.
In development set evaluations using patient-level out-of-fold predictions from 10-fold cross-validation, the ensemble model achieved an area under the receiver operating characteristic curve of 0.877, alongside an area under the curve of 0.865 in the independent external validation set. Complementary feature-contribution analysis using SHAP for the AdaBoost base classifier showed that liver enzymes and metabolic markers served as major contributors to model predictions.
External validation revealed an overprediction of absolute risk, demonstrating that the stacking ensemble model requires additional recalibration and broader validation in larger and more diverse pediatric populations before routine clinical deployment, while a web-based application was created to display individualized outputs.
The researchers noted that because the model was built entirely on routinely available clinical indicators—encompassing routine blood tests, anthropometric measurements, and biochemical markers—it offers an accessible framework for evaluating pediatric metabolic dysfunction-associated steatotic liver disease in outpatient settings. In particular, the authors concluded that this multicenter machine-learning approach may support future clinical identification of ultrasound-defined disease in overweight and obese children once further validation and recalibration are achieved.
Subject of Research: Machine learning model for ultrasound-defined pediatric MASLD screening
Article Title: Development and external validation of a stacking ensemble machine-learning model to identify ultrasound-defined MASLD in children with overweight or obesity
Article References: Yan, W., Shi, X., Hua, J., Zhao, B., Huang, C., Peng, L., Zhou, N., Hao, J., Bai, R., Liu, Q., & Li, X. (2026). Development and external validation of a stacking ensemble machine-learning model to identify ultrasound-defined MASLD in children with overweight or obesity. BMC Medical Informatics and Decision Making. https://doi.org/10.1186/s12911-026-03872-9
Image Credits: AI Generated
DOI: 10.1186/s12911-026-03872-9
Keywords: Machine Learning, Pediatric MASLD, Childhood Obesity, Abdominal Ultrasound, Clinical Prediction, Ensemble Learning, Medical Informatics, Pediatric Hepatology, Development, external, validation, stacking
News Source: Teresa Odom. (November 9, 2026). Machine Learning Model Identifies Ultrasound-Defined MASLD in Overweight Children. Scienmag.



