The Application Of Radiomics In Differential Diagnosis Of Peripheral Lung Adenocarcinoma And Peripheral Lung Squamous Cell Carcinoma

Objective: To explore the differential diagnosis ability of lung tissue based on lung CT enhanced images in peripheral lung adenocarcinoma and peripheral lung squamous cell carcinoma.Methods: A retrospective collection of 81 patients with peripheral lung adenocarcinoma and 69 patients with peripheral lung squamous cell carcinoma who underwent pulmonary enhancement CT examination from January 1,2018 to January 30,2019.The patient's CT image was imported into AK(Analysis-Kinetics,GE Healthcare,China)analysis software,and the region of interest(ROI)was delineated in the CT image.The computer automatically extracted 396 radiographic features and was screened by dimensional reduction.A key omics feature.Based on the 70% training group data,the characteristics obtained by the above dimensionality reduction are substituted into the multivariate logistic regression analysis to obtain the optimal logistic regression model.Finally,the independent 30% test group data is substituted into the constructed model to test the model performance.By plotting the receiver operating characteristic curve(ROC curve),the corresponding sensitivity,specificity and area under the curve and the corresponding95% confidence interval in the training and test groups are obtained respectively,and the decision is drawn simultaneously.The curve(decision curve analysis,DCA)evaluates the benefits that the model brings to patients.Results : After the dimension reduction,five key omics features are selected:Correlation?angle90?offset4,Inertia?angle90?offset7,HighGreyLevelRunEmphasis AllDi rection?offset1 SD,RunLengthNonuniformity?angle0?offset1,VolumeCC.The diagnostic sensitivity of the logistic regression model based on this was 0.667 and 0.762,respectively,in the training group and test group.The specificities were 0.839 and 0.600 respectively.The accuracy was 0.760 and 0.724,respectively.The area under the ROC curve(AUC)was 0.781(0.690-0.871)and 0.724(0.565-0.883),respectively.The decision curve shows that it has certain application value in clinical work.Conclusion: The radiomics model can distinguish peripheral lung adenocarcinoma from peripheral lung squamous cell carcinoma.Frankly speaking,the machine learning modelmay become a good non-invasive diagnostic method.