Machine learning models based on quantitative ultrasound and clinical indexes for predicting metabolic associated fatty liver disease
10.13929/j.issn.1672-8475.2025.06.006
- VernacularTitle:基于定量超声及临床指标机器学习模型预测代谢相关脂肪性肝病
- Author:
Xinge CAO
1
;
Yali ZHANG
;
Lizhuo JIA
;
Jianghong CHEN
;
Yi DONG
Author Information
1. 河北医科大学第一医院超声科,河北 石家庄 050000
- Publication Type:Journal Article
- Keywords:
fatty liver;
metabolic diseases;
ultrasonography;
machine learning
- From:
Chinese Journal of Interventional Imaging and Therapy
2025;22(6):394-399
- CountryChina
- Language:Chinese
-
Abstract:
Objective To observe the value of machine learning(ML)models based on quantitative ultrasound(QUS)and clinical indexes for predicting metabolic associated fatty liver disease(MAFLD).Methods Totally 298 patients underwent abdominal MR and QUS examinations were retrospectively enrolled,including 150 cases with and 148 cases without MAFLD.The patients were divided into training set(including 107 cases of MAFLD and 101 cases of non-MAFLD)and test set(including 43 cases of MAFLD and 47 cases of non-MAFLD)at a ratio of 7∶3.Features were selected using least absolute shrinkage and selection operator(LASSO)regression and logistic regression(LR),based on which predictive models were constructed using 6 ML classifiers,including Gaussian naive Bayes(GNB),LR,random forest(RF),support vector machine(SVM),extreme gradient boosting(XGBoost)and K-nearest neighbor(KNN),respectively.Then the receiver operating characteristic curves were drawn,and the area under the curve(AUC)and the Brier score were calculated to evaluate the predictive efficacy of the models.Results The elevated age,glutamic-pyruvic transaminase(GPT),glutamic-oxaloacetic transaminase(GOT),uric acid(UA),low-density lipoprotein cholesterol(LDL-C),controlled attenuation parameter(CAP),ultrasound-derived fat fraction(UDFF)and shear wave velocity(SWV),as well as blurred liver contour were all independent indicators for higher likelihood of MAFLD(all P＜0.05).The AUC and Brier score of XGBoost model in training set was 0.991 and 0.006,in test set was 0.973 and 0.069,respectively,both higher than those of other models,and decision curve analysis(DCA)indicated that XGBoost model had high net benefit.Conclusion ML models based on QUS and clinical indexes,especially XGBoost model had high efficacy for predicting MAFLD.
