Related Experiment Video
Updated: Aug 14, 2026
Optimized Analysis of In Vivo and In Vitro Hepatic Steatosis
Published on: March 11, 2017
An interpretable machine learning model based on routine clinical indicators for early identification of
Lili Zhuang1, Jinxing Dai1, Jiajia Wang2
1Department of Pediatrics, Siyang Hospital, Suqian, China.
Insights
A new machine learning model can predict non-alcoholic fatty liver disease (NAFLD) risk in obese children using routine tests. This tool aids in early screening and prioritizing referrals for further examination.
Area of Science:
- Pediatric Endocrinology
- Hepatology
- Medical Informatics
Background:
- Obesity in children increases the risk of non-alcoholic fatty liver disease (NAFLD).
- Current methods for NAFLD screening in obese children lack non-invasive, individualized tools.
- There's a need for tools to prioritize children with obesity for confirmatory ultrasonography.
Purpose of the Study:
- To develop and interpret a machine learning (ML)-based model for predicting ultrasound-defined NAFLD risk in obese children.
- To utilize routine clinical and biochemical indicators for NAFLD risk prediction.
- To create an interpretable model for clinical decision support.
Main Methods:
- A retrospective study of 439 obese children.
- Collected demographic, anthropometric, and laboratory data.
- Developed and compared six ML models (logistic regression, decision tree, random forest, XGBoost, LightGBM, SVM) using LASSO and Boruta algorithms for feature selection.
- Interpreted the optimal model using SHAP analysis.
Main Results:
- The LightGBM model demonstrated superior performance (AUC: 0.870, accuracy: 0.823).
- Key predictors identified include waist-to-height ratio, BMI, TyG-WC, serum uric acid, and TyG-BMI.
- The model accurately identified 46.9% of participants with ultrasound-defined NAFLD.
Conclusions:
- An interpretable ML model effectively predicts NAFLD risk in obese children using routine indicators.
- The LightGBM model shows potential as a non-invasive auxiliary tool for early NAFLD screening and referral prioritization.
- Further multicenter prospective studies and external validation are recommended for broader clinical application.
Background:
There is still a lack of a noninvasive, individualized tool to pre-screen children with obesity for the risk of non-alcoholic fatty liver disease (NAFLD) and to prioritize referral for confirmatory abdominal ultrasonography. In this study, we aimed to develop and interpret a machine learning (ML)-based model for predicting ultrasound-defined NAFLD risk among children with obesity using routine clinical and biochemical indicators.
Methods:
This single-center retrospective study included 439 obese children treated at Siyang Hospital between February 2021 and December 2025. Demographic, anthropometric, and routine laboratory indicators were collected, and ultrasound-defined NAFLD was used as the study outcome. Least absolute shrinkage and selection operator (LASSO) regression and the Boruta algorithm were applied to identify robust predictive features, and six ML models-logistic regression, decision tree, random forest, extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and support vector machine (SVM)-were developed and compared to determine the optimal model. Model performance was evaluated in the test set, and Shapley Additive exPlanations (SHAP) was used to interpret the final model.
Results:
A total of 439 obese children were included in the analysis, of whom 206 (46.9%) had ultrasound-defined NAFLD. Seven variables were selected to develop six ML models. Among these, the LightGBM model showed the best overall performance across multiple metrics, with comparable discrimination among the top models in the test set, with an area under the curve (AUC) of 0.870, accuracy of 0.823, precision of 0.828, sensitivity of 0.787, specificity of 0.855, and F1 score of 0.807. SHAP analysis further demonstrated that waist-to-height ratio contributed most to the model output, followed by body mass index (BMI), triglyceride-glucose index × waist circumference (TyG-WC), serum uric acid, and triglyceride-glucose index × body mass index (TyG-BMI).
Conclusions:
This study developed an interpretable ML-based model for predicting ultrasound-defined NAFLD risk in obese children using routine clinical and biochemical indicators. LightGBM showed the best overall predictive performance, with good calibration and potential clinical utility. It may serve as a noninvasive auxiliary tool to pre-screen children with obesity and prioritize referral for confirmatory abdominal ultrasonography, supporting early identification and risk assessment of ultrasound-defined NAFLD, although further multicenter prospective studies and external validation are still needed before broader clinical application.
