Whole lung CT radiomics combined with clinical and conventional CT features for distinguishing non-tuberculous mycobacterial pulmonary disease and pulmonary tuberculosis
10.13929/j.issn.1672-8475.2025.01.004
- VernacularTitle:全肺CT影像组学联合临床及常规CT特征鉴别非结核分枝杆菌肺病与肺结核
- Author:
Jie SHEN
1
;
Minlin YU
;
Xiaomei JIN
;
Lin ZHANG
;
Zeyang YU
;
Shicheng FENG
;
Ling WEN
Author Information
1. 南京医科大学附属脑科医院胸科院区放射科,江苏 南京 210009
- Publication Type:Journal Article
- Keywords:
tuberculosis,pulmonary;
mycobacterium infections;
tomography,X-ray computed;
radiomics
- From:
Chinese Journal of Interventional Imaging and Therapy
2025;22(1):16-21
- CountryChina
- Language:Chinese
-
Abstract:
Objective To investigate the value of whole lung CT radiomics combined with clinical and conventional CT features for differentiating non-tuberculous mycobacterial pulmonary disease(NTM-PD)and pulmonary tuberculosis(PTB).Methods Fifty-three NTM-PD(NTM-PD group)and 111 PTB(PTB group)patients diagnosed by mycobacteria culture were retrospectively collected.The patients were divided into training set(n=115,including 37 cases of NTM-PD and 78 cases of PTB)and test set(n=49,including 16 cases of NTM-PD and 33 cases of PTB)at the ratio of 7∶3.Patients'clinical and pulmonary CT manifestations of lesions were analyzed using univariate and multivariate logistic regression,and the independent impact factors for differentiating NTM-PD and PTB lesions were screened.The best radiomics features were extracted and screened based on whole lung CT.Based on independent impact factors,the best radiomics features and their combination,clinical-CT,radiomics and combined models were constructed with random forest,and the differential efficacy of each model was evaluated.Results Patients'age(OR=0.264),gender(OR=0.956),immunoglobulin G(OR=3.416),C reactive protein(OR=3.418)and bronchiectasis shown on CT(OR=0.285)were all independent impact factors for differentiating NTM-PD and PTB.Twelve best radiomics features were screened based on whole lung ROI.The AUC of combined model in training set and test set was 0.915 and 0.901,respectively,both higher than that of clinical model(AUC=0.832,0.801,Z=1.340,3.710,both P＜0.05)and radiomics model(AUC=0.877,0.821,Z=-2.520,-5.240,both P＜0.05).Conclusion Whole lung CT radiomics combined with clinical and conventional CT features could effectively distinguish NTM-PD and PTB.
