| Objective:To establish radiomics models of lung squamous cell carcinoma and adenocarcinoma by radiomics study of lung cancer.To validate the radiomics model on the discrimination of lung adenocarcinoma and squamous cell carcinoma applying the arterial phase and venous phase.Methods:(1)One hundred pathologically confirmed lung adenocarcinomas and one hundred pathologically confirmed lung squamous cell carcinoma from April 2015 to September 2017 were retrospectively collected as primary cohort.The age range is 16 to 85 years old,with a median age of 63 years old.There were 118 males and 82 females;103 smokers and 97 non-smokers;170 patients with stage I and II,and 30 patients with stage III and IV.All patients were treated with a Siemens dual-source CT scanner with a reconstruction layer thickness of 1 mm and a reconstruction layer spacing of 1 mm.The arterial phase and venous phase DICOM format images of the desired image enhancement scan were extracted from the PACS system of our hospital.(2)Manually segmentation of images using GE Healthcare’s Omini-Kineti software to extract CT radiomics features including tumor intensity,nodule volume,density,histogram,texture,and wavelet characteristics.Select the region of interest of the image,and sketch the target area.The hooks in the sketching process are as dense as possible,so as to accurately sketch.(3)Statistical analysis of the data was performed using SPSS 22.0 statistical software.Non-parametric tests(Mann-Whitney U test)and receiver operating curve(ROC)were used to compare CT values inside the Region of interest(ROI)to identify the lungs adenocarcinoma and lung squamous cell carcinoma.ROC curve analysis was used to calculate the area under the curve to identify lung adenocarcinoma and lung squamous cell carcinoma.The area under the curve was greater than 0.7.(4)Fisher discriminant analysis was performed to develop models.The generalization ability was evaluated with the cross-validation method.The accuracy is evaluated and the best method is screened by various test methods.This retrospective analysis was approved by the Ethics Committee of the Affiliated Hospital of Qingdao University.Results:(1)There was a significant difference in gender and smoking status between lung adenocarcinoma and lung squamous cell carcinoma.Females with lung adenocarcinoma were significantly more than women with lung squamous cell carcinoma(P < 0.05),and smokers with lung adenocarcinoma were significantly less than lung squamous cell carcinoma smokers(P <0.05).(2)A total of 398 parameters were extracted for radiomics analysis.In the non-parametric test(Mann-Whitney U test),there were 6 and 9 parameters in the arterial phase and venous phase,respectively,indicating significant differences between the groups(P < 0.05).However,the area under the curve was found to be less than 0.7(AUC < 0.7),indicating that the ROC curve and AUC did not have a good discriminating effect on the training set.(3)43 and 87 optimal features were screened in the arterial phase and the venous phase,and the optimal features were used to establish a model for the differentiation of lung squamous cell carcinoma and lung adenocarcinoma.It was found that,in the arterial phase,the identification accuracy of the initial grouping and the cross-validation group were 86.6% and 79.4%,respectively.In the venous phase,the identification accuracy of the initial and cross-validation groups was 97.0% and 83.8%,respectively.Conclusions:(1)Gender and smoking status are important predictors of histological classification of lung adenocarcinoma and lung squamous cell carcinoma.(2)The radiomics features are independent biomarkers used to identify lung squamous cell carcinoma and lung adenocarcinoma.(3)The radiomics model combined with radiomics features can distinguish between lung squamous cell carcinoma and lung adenocarcinoma.The radiomics model established by using arterial phase and venous phase has excellent diagnostic performance.(4)The radiomics model of the venous phase has the best predictive effect. |
PDF Full Text Request