Radiomics Study On Differential Diagnosis Of Benign And Malignant Pulmonary Nodules Based On Spectral CT

Objects:To explore the ability of radiomics features based on spectral CT to distinguish benign from malignant pulmonary nodules.Methods:A total of 242 patients with pulmonary nodules（61 benign and 181 malignant）from 2018 to 2020 in Shengjing Hospital of China Medical University were randomly divided into training set（n=61）and verification set（n=74）according to the proportion of 7:3.The 40-65Ke V images of traditional CT and spectral CT in arterial and venous phase were sketched by ROI and the radiomics features were extracted,and the spectral quantitative parameters(IC,NIC,CT_{40Ke V},λ_HU,Z_eff)of pulmonary nodules were measured.The traditional model based on clinical factors and spectral CT quantitative parameters（Model 1）were established by multi-factor Logistic regression analysis,14 different radiomics models were constructed based on the mono-energy images with different Ke V values and the features extracted from conventional images,and the model with the best discrimination efficiency was selected as the final radiomics model（Model 2）.An integrated model of radiomics features based on spectral CT optimal Ke V,clinical factors and quantitative parameters of spectral CT was established（Model 3）.The diagnostic efficiency of the model was evaluated by the receiver working characteristic（ROC）curve,and the accuracy,sensitivity,specificity and the area under the ROC curve（AUC）were calculated.Draw the calibration curve and compare the fit degree of the model by Hosmer-Lemeshow test;draw the decision curve to compare the clinical benefits of the model.Results:The differential diagnosis of benign and malignant pulmonary nodules was made based on 40-65Ke V and conventional images of spectral CT,in which the radiomics models established based on arterial and venous 65Ke V images were in the training set（arterial phase AUC=0.94[95%CI:0.90,0.97],venous phase AUC=0.92[95%CI:0.87,0.97]）and verification set（arterial phase AUC=0.92[95%CI:0.84,0.98],venous phase AUC=0.88[95%CI:0.79,0.95]）was higher,and there was no significant difference in diagnostic efficiency between the two stages in training set（0.58）and verification set（0.51）.The radiomics model established by the combination of arterial and venous 65Ke V images of spectral CT in the training set and verification set（AUC=0.96[95%CI:0.93,1.00],0.96[95%CI:0.91,1.00]）is superior to the traditional model（AUC=0.88[95%CI:0.82,0.94],0.86[95%CI:0.77,0.95]）in the differential diagnosis of benign and malignant pulmonary nodules.Compared with the radiomics model,the diagnostic value of the integrated model based on the radiomics features extracted from arterial and venous65Ke V images,clinical factors and spectral CT quantitative parameters in the training set and verification set is slightly improved（AUC=0.97[95%CI:0.94,1.00],0.97[95%CI:0.93,1.00]）.Conclusion:In the 40-65Ke V and conventional images of spectral CT during arterial and venous phase,the radiomics model based on arterial and venous phase 65Ke V images of spectral CT is of high diagnostic value in differentiating benign from malignant pulmonary nodules,and the integrated model based on clinical and spectral CT quantitative parameters and radiomics features is obviously superior to clinical and spectral CT quantitative parameter models in qualitative diagnosis of pulmonary nodules.The radiomics based on spectral CT can provide more quantitative diagnostic information for the qualitative diagnosis of pulmonary nodules,which is helpful to further reveal the heterogeneity of tumors.