Abstract
PURPOSE: This study aims to develop and validate a clinical-radiomics fusion model that integrates clinical characteristics with CT-derived radiomic features for the rapid and accurate prediction of drug resistance in patients with cavitary pulmonary tuberculosis (TB). PATIENTS AND METHODS: A total of 231 patients with microbiologically confirmed cavitary pulmonary TB were retrospectively enrolled and divided into a drug-resistant TB group (n=89) and a drug-sensitive TB group (n=142) based on drug susceptibility testing. Radiomics features were extracted from CT images, and a radiomic signature was constructed following stability assessment, dimensionality reduction, and least absolute shrinkage and LASSO regression. Clinical predictors with significant statistical differences were identified. Independent clinical, radiomic, and combined clinical-radiomics fusion models (nomogram) were developed using eight machine learning algorithms, and their performances were compared. Models' performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. RESULTS: Four clinical variables (C-reactive protein, diabetes history, alanine aminotransferase, and blood calcium levels) and a radiomics signature comprising 14 key features were selected as final predictors. The fusion model achieved AUCs of 0.861 in the training set and 0.884 in the test set, outperforming both the standalone clinical and radiomic models. Decision curve analysis demonstrated that the fusion model provided higher clinical net benefit across a wide range of threshold probabilities. CONCLUSION: The proposed clinical-radiomics fusion model enables accurate prediction of drug resistance in cavitary pulmonary TB, supporting the optimization of initial treatment strategies and promoting the implementation of precision medicine in TB management.