Abstract:

Objective: To explore the value of combining CT-based radiomics with clinical features in distinguishing pulmonary tuberculosis (PTB) from nontuberculous mycobacteria pulmonary disease (NTM-PD). Methods: A retrospective analysis was conducted on clinical data and CT images of NTM-PD and PTB patients confirmed with culture from The First Affiliated Hospital of Xinxiang Medical University from January 1, 2019, to March 31, 2023. Based on the results of bacterial culture, all patients were divided into the PTB group (58 cases) and the NTM-PD group (75 cases). Clinical features were analyzed, and statistically significant features were used to construct a clinical model. CT images were used to study cavitary lesions, with a total of 200 lesions included in the study. The lesions were randomly divided into a training set and a testing set in a 7∶3 ratio. A logistic regression classifier was used to construct a radiomics model. A combined model was built by integrating radiomics features and clinical features. The diagnostic performance of the models in the training and testing sets was evaluated by sensitivity, specificity, accuracy, receiver operating characteristic (ROC) curve, area under the curve (AUC), and calibration curve. Results: Univariate analysis showed that there was statistically significant difference in age between the PTB group (median (quartile) age 45 (26, 66) years) and the NTM-PD group (median (quartile) age 63 (54, 70) years)(Z=-3.184, P<0.001). There was also statistically significant difference in BMI between the PTB group (19.95±2.83) and the NTM-PD group (18.78±2.59)(t=2.469, P=0.015). The proportions of patients with positive interferon-gamma release assays (IGRA) results were significantly different between the PTB group (55 cases, 73.33%) and the NTM-PD group (16 cases, 27.59%)(χ²=27.505, P<0.001). Multivariate analysis showed that age (OR=0.969, P=0.004) and IGRA (OR=6.026, P<0.001) were independent predictive factors for distinguishing PTB from NTM-PD. The AUC values of the clinical model in the training and testing sets were 0.832 (95%CI: 0.765-0.899) and 0.800 (95%CI: 0.689-0.911), respectively. The AUC values of the radiomics model in the training and testing sets were 0.974 (95%CI: 0.952-0.996) and 0.939 (95%CI: 0.877-1.000), respectively. The AUC values of the combined model in the training and testing sets were 0.993 (95%CI: 0.986-1.000) and 0.995 (95%CI: 0.985-1.000), respectively. Conclusion: The combined model incorporating clinical features and radiomics features is a non-invasive, convenient, and rapid diagnostic method that can effectively distinguish PTB from NTM-PD.

Key words: Tuberculosis,pulmonary, Mycobacterium infections, Diagnosis,differential, Tomography,X-ray compute, Radiomics