Machine learning and radiomics for predicting therapeutic efficacy in newly diagnosed sputum-negative pulmonary tuberculosis: a retrospective study.

Background: Early assessment of tuberculosis progression and the efficacy of anti-tuberculosis drugs is crucial for improving disease cure rates. This study aimed to predict the therapeutic efficacy of newly diagnosed sputum-negative but bronchial alveolar lavage fluid (BALF)-positive pulmonary tuberculosis patients after intensive therapy. Methods: We collected data from 255 patients (178 and 77 in center 1 and 2, respectively) diagnosed with newly diagnosed sputum-negative but BALF-positive pulmonary tuberculosis. Based on imaging and clinical follow-up results, the patients were divided into progression and improvement groups. Radiomics features were extracted from five computed tomography (CT) signs, and feature selection was performed using Pearson correlation analysis and the Least Absolute Shrinkage and Selection Operator (LASSO). Three machine learning models (random forest (RF), support vector machine (SVM), and logistic regression (LR)) were then constructed. Predictive performance was evaluated using receiver operating characteristic (ROC) curves, F1 scores, and Delong tests. Results: A total of 118 radiomics features were used to construct three models that demonstrated good performance. In the training and test cohorts, the SVM model achieved area under the curves (AUCs) of 0.917 and 0.858, and F1 scores of 0.808 and 0.755, respectively. The RF model showed the highest predictive performance with AUCs of 0.996 and 0.824, and F1 scores of 0.982 and 0.832. The LR model achieved AUCs of 0.927 and 0.808, and F1 scores of 0.867 and 0.747. Conclusions: Machine learning models based on radiomic features extracted from various CT signs demonstrate potential for predicting the therapeutic efficacy in newly diagnosed pulmonary tuberculosis patients after intensive therapy, providing effective guidance for subsequent treatment.