Related Experiment Videos
The application of explainable machine learning for the identification of MASLD in non-obese populations
Huiyan Yu1, Xiuli Han2, Nayi Zhang2
1Department of Critical Care Medicine, Qingdao Central Hospital, University of Health and Rehabilitation Sciences (Qingdao Central Hospital), Qingdao, China.
Background:
The diagnosis of metabolic dysfunction-associated steatotic liver disease (MASLD) in non-obese individuals presents a significant challenge, and its early identification is of considerable public health importance. This study aimed to develop and validate interpretable machine learning models using routine clinical parameters to identify MASLD in this specific population.
Methods:
Data from 768 non-obese subjects were analyzed. The dataset was randomly split into training (n = 615, 80%) and testing (n = 153, 20%) sets using stratified sampling. LASSO regression was performed within the training set for feature selection. Eight machine learning models were constructed in the training set: Linear Discriminant Analysis (LDA), Gradient Boosting Machine (GBM), Neural Network (NN), Logistic Regression (LR), Support Vector Machine (SVM), Naïve Bayes (NB), Random Forest (RF), and Decision Tree (DT). Model performance was evaluated using AUC, calibration analysis and decision curve analysis (DCA). SHAP analysis was applied to interpret the optimal models.
Results:
LASSO regression identified eight factors: apolipoprotein B, triglycerides, fasting insulin, fasting plasma glucose, waist circumference, age, alkaline phosphatase, and systolic blood pressure. On the test set, GBM achieved the highest AUC of 0.779 (95% CI: 0.695-0.863), followed by NN (0.777) and LDA (0.771). Both GBM and LDA demonstrated acceptable calibration (Hosmer-Lemeshow p > 0.05) and clinical utility. SHAP analysis revealed that fasting insulin was the most influential indicator for LDA, while triglycerides dominated GBM case identification.
Conclusion:
This study developed interpretable machine learning models (LDA and GBM) for identifying MASLD in non-obese individuals using readily available clinical parameters. The models demonstrated acceptable discrimination and calibration, with SHAP analysis highlighting the central roles of insulin resistance and dyslipidemia in the pathogenesis of non-obese MASLD.