A Combined Clinical-radiomics Model Based On Energy Spectrum Enhanced CT To Discriminate Between Isolated Solid Benign Pulmonary Nodules And Pulmonary Adenocarcinoma

ObjectiveThe purpose of this study was to evaluate the role of different regional and Energy spectrum CT radiomics features in identifying benign lesions and lung adenocarcinoma(LAC)manifesting as pulmonary solitary solid nodule(PSSN).We also selected the best features to construct a combined clinical-radiomics model and develop a nomogram to enhance the discrimination between benign and malignant isolated solid lung nodules.MethodsThis study retrospectively analyzed 288 patients with benign pulmonary nodules(140 cases)and adenocarcinoma nodules of the lung(148 cases)who underwent consultation and were clinically and pathologically confirmed at the Second Affiliated Hospital of Nanchang University between May 2021 and October 2022,randomly divided into a training set(201 cases)and a validation set(87 cases)according to a ratio of 7:3.Basic clinical information and initial Energy spectrum CT images were collected from all patients.Post-processing techniques were used to obtain quantitative CT parameters of all lesions,including: Normalized Iodine Concentration(NIC),Normalized effective atomic number(NZeff),and Slope of Spectral Attenuation Curves(λHU).The tumor body and peri-tumor 5 mm region of interest were outlined layer by layer on two sets of single-energy images(40 KEV and 100 KEV)in the enhanced venous phase,and the radiomics features were extracted.The intraclass Correlation Efficient(ICC),pearson/spearman correlation analysis,Least Absolute Shrinkage and Selection Operator(LASSO)and variance inflation factor(VIF)algorithms were then used for downscaling and screening of radiomics features.Univariate and multifactor analyses were used for screening of clinical-visual radiological features.Eight models were constructed using logistic regression for clinical radiology,40 KEV intra-nodal,40 KEV peri-nodal,40 KEV intra-and perinodal,100 KEV intra-nodal,100 KEV peri-nodal,100 KEV intra-and peri-nodal,and clinical + 100 KEV intra-and peri-nodal,respectively.Receiver Operating Characteristics(ROC),Area under curve(AUC),Delong test,calibration curve and Hosmer-Lemeshow test were used for model performance evaluation.ResultsUnivariate and multifactorial risk analyses showed that three clinical radiology characteristics,namely age,serum CEA level,and pleural retraction,were independent risk factors for identification,and the clinical model constructed based on these three characteristics had AUC=0.72,95% CI 0.65-0.79 in the training set and AUC=0.70,95% CI 0.58-0.81 in the validation set.all six radiomics models performed better than the clinical model: AUC=0.84,95% CI 0.78-0.89 in the training set and AUC=0.84,95% CI 0.76 in the validation set.Therefore,combining age,serum CEA level,pleural retraction and 100 KEV intra-and peri-nodal radiomics features to construct a combined model had the best predictive performance among all models: training set AUC=0.84,95% CI 0.78-0.89,validation set AUC=0.84,95% CI 0.76-0.92.The Hosmer-Lemeshow test and calibration curve indicated that the combined model fitted well,and the clinical decision curve reflected the high clinical utility value of the combined model.Conclusion(1)Radiomics features perform well in the application of identifying isolated solid lung benign nodules and pulmonary adenocarcinoma,outperforming traditional clinical factors,quantitative parameters of Energy spectrum CT and visual radiological features.(2)Good performance of radiomics features in both intra-tumor and peri-tumor regions in the differentiation of benign and malignant isolated solid lung nodules,with better performance of 100 KEV radiomics features than 40 KEV radiomics features.(3)The combined model combining clinical factors,radiomics features possesses better performance and improves the discrimination of isolated solid lung benign nodules and pulmonary adenocarcinoma.