An Ultrasound-Based Machine Learning Model for Differentiating IgG4-Related Sialadenitis and Primary Sjögren's Syndrome.
Authors
Affiliations (3)
Affiliations (3)
- Department of Ultrasound, The First Affiliated Hospital of Xiamen University, School of Medicine, Xiamen University, Xiamen, China. Electronic address: [email protected].
- Department of Ultrasound, The First Affiliated Hospital of Xiamen University, School of Medicine, Xiamen University, Xiamen, China.
- Department of Ultrasound, The First Affiliated Hospital of Xiamen University, School of Medicine, Xiamen University, Xiamen, China. Electronic address: [email protected].
Abstract
To establish and validate an interpretable machine learning (ML) model based on ultrasound (US) scoring system for differentiating immunoglobulin G4-related sialadenitis (IgG4-RS) from primary Sjögren's syndrome (pSS). A total of 263 patients with IgG4-RS or pSS were allocated to training (n = 184) and validation (n = 79) cohorts. Class imbalance was addressed using the synthetic minority oversampling technique. Optimal features were selected via least absolute shrinkage and selection operator (LASSO) regression, with different predictive models constructed using logistic regression alongside eight ML algorithms. Model performance was assessed through multiple metrics, including the area under the curve (AUC), while feature contributions were quantified using the Shapley Additive explanations (SHAP) algorithm. LASSO regression identified eight key features: sex, dry mouth, dry eyes, parotid gland enlargement, submandibular gland enlargement, parotid gland ultrasound (PGUS) scores, submandibular gland ultrasound (SMGUS) scores, and submandibular gland vascularity (SMGV) scores. Among the developed models, the light gradient boosting machine model performed best, achieving an AUC of 0.955, accuracy of 93.7%, sensitivity of 96.5%, and specificity of 86.4% in the validation cohort; decision curve analysis confirmed its superior clinical utility over others. Females, low SMGUS scores, high PGUS scores, absence of submandibular gland enlargement, and low SMGV scores were identified as top five pSS predictors using the SHAP analysis. The interpretable ML model based on US scoring system offers an accurate, non-invasive tool to differentiate IgG4-RS from pSS.