Exploration Of Imaging Markers Of Lung Adenocarcinoma Evolution And Construction Diagnostic Model Of Solitary Pulmonary Nodule Based On Quantitative CT Radiomics

Objective:To evaluate and analyze the radiomics information solitary pulmonary nodules quantified by CAD,clinical data and pathological diagnostic information,and to discover the imaging markers of lung adenocarcinoma evolution and to build the clinical diagnostic model of solitary pulmonary nodules.To explore whether the quantitative radiomics model can predict the benign or malignant diagnosis of the solitary pulmonary nodules and the aggressive grade of lung adenocarcinoma,so as to guide the clinical treatment strategy of the pulmonary nodules.Methods:In this retrospective study,which included 455 patients with definite pathological diagnosis of solitary pulmonary nodules in our institution from Sep 2016 to Sep 2018,the clinical data,imaging data and pathological diagnosis data were collected and analyzed.In this study,the stage of lung cancer was pT1abcNOMO,stage IA,and the CT image of benign lesions was SPN with no greater than 3.0cm diameter.The data included pathological diagnosis,preoperative CT diagnosis,gender,age,smoking history,family history of tumor,previous history,contact history.The quantitative image information(location,multiple primary,diameter,average density)and radiomics information(surface area,volume,MASS,irregularity,proportion of ground glass,proportion of fat,proportion of emphysema,proportion of calcified volume,COPD950,COPD910,proportion of emphysema,proportion of pleura,number of related vessels,vascular density,vascular tortuosity)of SPNs were provided by CAD.First,to evaluate the accuracy of the quantified image information and radiomics information provided by CAD,the diagnostic accuracy of SPN was compared between conventional CT diagnosis and CAD diagnosis platform,and the invasion grade of adenocarcinoma and SPN diameter were stratified.Second,the independent predictors of aggressive grade of lung adenocarcinoma were estimated with multivariate analysis,then the aggressive grade prediction model was built.Third,the predictors of malignancy were estimated with multivariate Logistic regression analysis,then the malignant prediction model was established.Last,ROC curve was drawn to evaluate the diagnostic ability of predictive markers and the models.Results:Grouped according to the invasive grade of lung adenocarcinoma,13.8%patients were in AAH group,1 1.4%patients were in AIS group,12.5%patients were in MIA group,37.4%patients were in IA group,24.8%patients were in BL group.53.1%of the nodules were malignant,and 46.9%were benign.Chi-squared Test verified that there was a statistical difference in the diagnostic accuracy of pulmonary nodules between CAD and conventional CT.And CAD was higher than conventional CT.The CAD diagnostic accuracy rate was 81.0%,the diagnostic error rate was 3.5%,among which the under-diagnosed was 0.6%,the overdiagnosis was 2.9%,the uncertainty was 15.5%.The conventional CT diagnostic accuracy was 63.7%,the diagnostic error was 12.0%,the under-diagnosed was 2.6%,the overdiagnosis was 9.4%,the uncertainty was 24.3%.Stratification according to the diameter of SPN(d≤10mm,10mm<d 20mm,and d>20mm),the diagnostic accuracy of conventional CT was 39.3%,64.4%,and 87.5%respectively,and the diagnostic accuracy of CAD was 65.2%,79.7%,and 98.2%,respectively.The diagnostic accuracy of CAD was higher than that of conventional CT(P<0.05).In multivariate ordered logistic regression analysis,12 factors related to the aggressive stage of lung adenocarcinoma(P<0.05).Among the 12 factors volume of SPN,surface area,irregular,Irregularty,average diameter,diameter,mean density,COPD910,fat ratio,number of related vascular,pleural proportion and fractal dimension are risk factors for lung adenocarcinoma aggressive stages,and the emphysema%around the SPN is a protective factor.In ROC curve analyses,the surface area of 8.05 cm2 regarded as the critical value of the diagnosis of IA,its AUC is 86.4%,the sensitivity is 82.4%and the specificity is 81.1%,and the SA/V of 8.92 is regarded as the critical value of the diagnosis of IA,the AUC is 73.5%,the sensitivity is 68.2%and the specificity is 68.1%,and the MASS of 439.02 mg is regarded as the critical value of the diagnosis of IA,the AUC is 80.1%,the sensitivity is 80.6%and the specificity is 69.1%,and the number of related vascular of 7 is regarded as the critical value of the diagnosis of IA,the AUC is 79.2%,sensitivity is 71.2%and specificity 78.2%,and the fractal dimension of 0.731 is taken as the critical value in the diagnosis of MIA,the AUC was 85.4%,its sensitivity is 74.9.1%,specificity is 86.4%,and the fractal dimension of 0.860 is taken as the critical value for the diagnosis of IA,its AUC is 91.1%,sensitivity is 81.2%and specificity is 89.5%.Logistic regression analysis showed that 14 radiomics factors[Volume(OR:1.08),area(OR:1.03)SA/V(OR:1.308),max diameter(OR:1.314),mean diameter(OR:1.64),MASS(OR:2.0),COPD950(OR:1.104),COPD910(OR:1.063),Emphysema%(OR:0.877),irregularity(OR:3.632),fat ratio(OR:1.110),Vessel Number(OR:1.03),Vessel Intensity(OR:0.999)and the fractal diameter(OR:2.029)]were independent predictors of malignancy in patients with SPN(P<0.05).The area under the ROC curve for our model was 0.8753(95%CI:0.8376-0.9131).Conclusion:1.The diagnostic accuracy of CAD for SPN is higher than that of conventional CT examination,and overdiagnosis is more than underdiagnosis in the diagnostic errors of CAD and conventional CT diagnosis;CAD "digital lung" can provide quantitative imaging information of SPN.2.Imaging and radiomics factors:SPN irregularity,SA/V,MASS,related vascular numbers,fat ratio,fractal dimension,and surface area can be used as imaging markers for lung adenocarcinoma invasive grade diagnosis(AAH,AIS,MIA,IA).3.Model included 12 quantifiable factors(family history,multiple primary lesions,olume,area,SA/V,the average diameter,COPD910,Emphysema,irregularity,fat ratio,density of the related vascular,pleural ratio)can be used in prediction of the invasive grade of lung adenocarcinoma(AAH,AIS,MIA,IA).4.14 radiomics factors(Volume,area,SA/V,max diameter,mean diameter,MASS,COPD950,COPD910,Emphysema%,irregularity,fat ratio,vessel Number,vessel intensity and the fractal diameter can be used as imaging markers for diagnosis of SPN malignant or benign.5.The prediction model including 6 radiomics factors is accurate and sufficient to estimate the malignancy of SPN.