The application of the radiomic-clinical model based on SHAP-XGBoost method for differentiating pulmonary tuberculosis from

Purpose: Diagnosing pulmonary tuberculosis (PTB) in children remains challenging due to its often non-specific clinical presentation. Early detection enables timely treatment initiation. This study aimed to develop and validate a predictive model to distinguish PTB from Methods: Lung CT images were obtained from 52 children with PTB and 80 with SPP, from which 1,023 radiomic features were extracted. The data were randomly split into training and test sets at a 7:3 ratio. Variance threshold, univariate analysis, and LASSO regression were applied sequentially to select the key radiomic features. Subsequently, an XGBoost model integrating clinical factors and radiomic features was constructed. The SHAP method was used to interpret the model and its decision process. Results: After feature selection, six radiomic features and three clinical features were retained for model construction. The combined radiomic-clinical model outperformed both the clinical-only and radiomic-only models. Its AUC was 0.981 in the training cohort and 0.897 in the test cohort. The clinical model achieved AUCs of 0.955 and 0.784, and the radiomic model achieved 0.868 and 0.865, respectively. SHAP summary and waterfall plots illustrated the contribution of each feature to the predictions. Conclusion: The SHAP-XGBoost radiomic-clinical model accurately distinguished PTB from SPP in children. It may offer a non-invasive and efficient tool to support clinical decision-making.