关键词: 结核,肺, 影像组学, 体层摄影术,X线计算机

Abstract:

Objective: To construct a model based on CT imaging radiomics combined with clinical features to predict the drug resistance of active pulmonary tuberculosis. Methods: The study included 234 patients with pulmonary tuberculosis admitted to The First Affiliated Hospital of Xinxiang Medical University from January 1, 2020 to December 31, 2022. Based on drug resistance status, the patients were divided into two groups: 88 cases in the drug-resistant group and 146 cases in the drug-sensitive group. They were then randomly assigned to a training set and a testing set in a ratio of 7∶3.Volume of interest (VOI) delineation was performed on the lesions, and radiomics features were extracted. Feature selection was conducted using the minimum redundancy maximum relevance (MRMR) and least absolute shrinkage and selection operator (LASSO) methods. Logistic regression was employed to establish clinical models and radiomics models separately. Subsequently, the selected optimal radiomics features, statistically significant clinical features, and CT features were combined to construct a joint model. The diagnostic performance of each model was evaluated using the area under the receiver operating characteristic curve (AUC). Results: Among the patients with drug resistance, there were 48 primary cases (54.55%) and 40 retreatment cases (45.45%). The detection rate of tree-in-bud sign was 69.32% (61/88). In the drug-sensitive group, there were 131 primary cases (89.73%) and 15 retreatment cases (10.27%). The detection rate of tree-in-bud sign was 81.51% (119/146). The clinical and CT feature analysis of patients in the drug-resistant and drug-sensitive group showed that there were statistically significant differences in treatment history (χ²=37.796, P<0.001) and tree-in-bud sign (χ²=4.595, P=0.032) between the two groups. Regarding CT findings analysis, the interobserver agreements between two physicians were good for the observation of nodules and satellite lesions, calcified nodules, consolidation, fibrotic bands, bronchial dilation, and tree-in-bud sign (Kappa coefficients were 0.757, 0.784, 0.818, 0.777, 0.863, and 0.781, respectively). A total of 14 radiomics features were selected as predictive indicators using the MRMR and LASSO methods to construct the prediction model. The AUC of the clinical model in the training set and testing set were 0.760 (95%CI: 0.687-0.834) and 0.820 (95%CI: 0.704-0.937), respectively. The AUC of the radiomics model in the training set and testing set were 0.822 (95%CI: 0.758-0.885)and 0.845 (95%CI: 0.744-0.947), respectively. The AUC of the combined model in the training set and testing set were 0.878 (95%CI: 0.823-0.932) and 0.888 (95%CI: 0.788-0.987), respectively. Conclusion: The radiomics model exhibited higher diagnostic performance than the clinical model, while the combined model showed the best diagnostic performance in both the training and testing sets.

Key words: Tuberculosis, pulmonary, Radiomics, Tomography, X-ray computed