Preliminary Study Of CT Radiomics To Differentiate Between Tuberculosis,Adenocarcinoma And Non-Tuberculous Infectious Lesions Presenting As Solid Pulmonary Nodules Or Masses

Part 1 Radiomics based on contrast-enhanced CT for differentiating between pulmonary tuberculosis and pulmonary adenocarcinoma presenting as solid nodules or massesObjective: The preoperative differentiation between pulmonary tuberculosis(PTB)and pulmonary adenocarcinoma(PAC)presenting solid pulmonary nodules or masses is a matter of patient management.This study investigated the incremental value of radiomics based on contrast-enhanced CT(CECT)in identifying PAC from PTB presenting as solid nodules or masses and established the best radiomics model.Materials and methods: In total,128 lesions(from 123 patients)were retrospectively analysed and randomly divided into training and test datasets in a 7:3 ratio.A subjective image model was developed using independent predictors of subjective image features.The best radiomics features based on non-contrast-enhanced CT(NCECT)and CECT were screened using the correlation coefficient method,univariate analysis,and the least absolute shrinkage and selection operator(LASSO)and was used to build the NCECT radiomics model and the CECT radiomics model.Finally,the established NCECT radiomics model and the CECT radiomics model were combined to construct a combined radiomics model.In addition,the diagnostic ability of three radiologists and one respiratory physician was assessed.Receiver operator characteristic(ROC)curves and calibration curves were used to assess the predictive accuracy of the models.Area under the curve(AUC)was used to assess the performance of the prediction model.Decision curve analyses(DCA)were also performed to assess whether the radiomics model could be applied in clinical practice.Results: Univariate and multivariate analyses showed that the spiculation,cavity and air bronchogram were independent predictors for identifying PTB and PAC.The subjective image model was created using these three features.The CECT radiomics model(training dataset: AUC=0.933;test dataset: AUC=0.881)outperformed the NCECT radiomics model(training dataset: AUC=0.861;test dataset: AUC=0.756)and subjective image model(training dataset: AUC=0.760;test dataset: AUC=0.611).The combined radiomics model(training data set: AUC=0.948;test data set: AUC=0.917)had the highest diagnostic efficacy,outperforming respiratory physicians and young and middle-aged radiologists.Conclusions: The CECT radiomics model is of greater value than the NCECT radiomics model in identifying PTB and PAC that present as solid nodules or masses.The combined radiomics model had the highest diagnostic efficacy and had the potential to aid clinical diagnosis,particularly for clinicians and young and middle-aged radiologists.Part 2 Radiomics based on contrast-enhanced CT for differentiating tuberculosis from nontuberculous infectious lesions presenting as pulmonary solid nodules or massObjective: To compare the value of contrast-enhanced CT(CECT)and non-contrast-enhanced CT(NCECT)radiomics models in differentiating pulmonary tuberculosis(PTB),which presents as a solid pulmonary nodules or masses,from non-tuberculous infectious lesions(NTIL),and to establish a stable and reliable combined model.Materials and methods: This study was a retrospective analysis of 101 lesions from 95 patients,49 lesions(from 45 patients)from the PTB group,and 52 lesions(from 50 patients)from the NTIL group.Lesions were randomly divided into training and test datasets in a 7:3 ratio.The conventional imaging model was constructed using conventional imaging features.Consistency testing,correlation analysis,the gradient boosting decision tree(GBDT),and logistic regression were used to screen radiomics features and construct the NCECT radiomics model and the CECT radiomics model.Finally,a combined model was constructed using a combination of the CIM,the NCECT radiomics model,and the CECT radiomics model.In addition,three radiologists were brought in for independent evaluation.The area under the receiver operating characteristic curve(AUC)was used to assess the differential diagnostic performance of each model.Results: Univariate analysis showed that only the difference in cavity was significant(P=0.001).The conventional imaging model was constructed using all conventional imaging features.The CECT radiomics model(training: AUC=0.874;test: AUC=0.796)outperformed the conventional imaging model(training: AUC=0.792;test: AUC=0.708),the NCECT radiomics model(training: AUC=0.835;test: AUC=0.704),and the three radiologists.The combined model(training: AUC=0.922;test: AUC=0.833)had the best diagnostic efficacy in both the training and test datasets.Conclusions: Radiomics helped to differentiate PTB from NTIL presenting as solid pulmonary nodules or masses,and CECT may be a better choice.The combined model obtained the best diagnostic efficacy and may outperform expert radiologists.