Preliminary Study On Prediction Of EGFR Mutation In Peripheral Lung Adenocarcinoma By Radiomics Based On 2D And 3D CT Features

Objective: Using the method of radiomics,we constructed the Normogram of imaging omics to explore its efficacy in predicting EGFR gene mutation in peripheral lung adenocarcinoma.Materials and methods:Collected from January 2018 to 2020 2 menstrual pathological verification with complete clinical and imaging data parallel immunohistochemical detection of 209 cases of peripheral lung adenocarcinoma,using one method 209 patients were divided into exercise group(146 cases)and the validation group(63 cases),in accordance with the EGFR mutation is divided into EGFR positive group and negative group,all of the patients were performed CT dynamic enhancement,from each patient CT images to extract 396 radiomics arterial characteristics,the ITK-SNAP software manual sketch each patient arterial image maximum level of the tumor,Thus,the region of interest(ROI)of 2D image is obtained.The ITK-SNAP software was used to manually sketch each layer of the tumor in the arterial phase image of each patient to reconstruct the 3D image and obtain the Region of Interest(ROI)of the 3D image.The software A.K was used to quantitatively extract the radiomics features from the delineated ROI.Lasso(minimum absolute contraction operator method)and MRMR(minimum redundancy and maximum correlation algorithm)were used to carry out the feature screening,and the significant features were selected and the radiomics labels were established in the training group.Logical regression analysis model was established by using the selected significant features and the multivariate Logistic regression analysis method,and the efficacy and clinical application value of 2D and 3D were evaluated by using the model.ROC(Receiver Operating Curve)and AUC(Area under Curve)were used to evaluate the clinical practical value of the constructed model.DCA(Decision Curve Analysis)was used to evaluate the clinical efficacy of the radiomics Normogram in the training group and the validation group.The larger the value of AUC,the better the classification effect of the model.The clinicopathological features included gender,age,smoking history,TNM stage and CT signs.These CT signs included vascular cluster sign,pleural depression sign,air bronchial sign,maximum tumor diameter(Dmax),lobulation sign,vacuole sign,and burr sign.According to the EGFR mutation status,the patients were divided into EGFR mutant group and wild-type group.The differences of EGFR mutation status in clinicopathological characteristics were observed by Wilcox-On rank sum test or Fisher test.Univariate and multivariate logistic regression analysis was used to analyze the EGFR gene mutation status and clinicopathological characteristics,and the clinical model was established.If p<0.05 means the difference is statistically significant.Results: In this study,there were 117 cases(117/209,56.0%)in the EGFR mutant group and 92 cases(92/209,44.0%)in the EGFR wild-type group.EGFR mutation status was significantly different in gender,smoking history,and patient TNM stage(P <0.05;0.05).The EGFR mutant group was more common in women who did not smoke.In2 D,multivariate logistic regression analysis showed that gender,tumor clinical stage,pleural depression and burr sign were independent risk factors for predicting EGFR gene mutation status in peripheral lung adenocarcinoma.In 2D,multivariate logistic regression analysis showed that gender,tumor clinical stage,burr sign and vascular cluster sign were independent risk factors for predicting EGFR mutation status in peripheral lung adenocarcinoma.In the training group(2D AUC,0.81;95% CI,0.74-0.88;3D AUC,0.72;95% CI,0.64-0.81;2D+3D AUC,0.96;95% CI,0.93-0.99)and validation group(2D AUC,0.80;95% CI,0.69-0.91;3D AUC,0.71;95% CI,0.58-0.85;2D+3D AUC,0.95;95%CI,0.90-1.00),it can be seen that the relevant radiomics labels established by the radiomics features have a good predictive effect.In the training group,the radiomics labels of 2D imaging Normogram included gender,TNM stage,burr sign and pleural depression sign.The radiomics labels of the 3D radiomics Normogram included gender,TNM stage,burr sign and vascular cluster sign.The radiomics labels of 2D+3D radiomics Normogram included smoking history,burr sign and vascular cluster sign.Compared with the clinicopathologic feature model,the imaging omics label showed better diagnostic efficacy(2D AUC,0.81;95% CI,0.74-0.88;3D AUC,0.72;95% CI,0.64-0.81;2D+3D AUC,0.96;95%CI,0.93-0.99)was higher than that of the clinical model(2D AUC,0.72;95% CI,0.64 to 0.80;3D AUC,0.72;95% CI,0.64 to 0.80;2D+3D AUC,0.70;95%CI,0.62-0.78).Radiomics Normogram also had the best predictive performance in the validation group(2D AUC,0.80;95% CI,0.69 to 0.91;3D AUC,0.71;95% CI,0.58 to 0.85;2D+3D AUC,0.95;95%CI,0.90-1.00).The prediction efficiency of 2D model was higher than that of 3D model,and 2D,3D and 2D+3D models were higher than that of clinical model(2D AUC,0.71;95% CI,0.58 to 0.84;3D AUC,0.69;95% CI,0.55 to 0.83;2D+3D AUC,0.64;95%CI,0.51-0.76)was also higher than 2D and 3D Normograms.In the training group and the verification group,the imaging nomogram showed better correction efficiency.Conclusion: Radiomics based on enhanced CT images can provide a method for predicting EGFR mutation status.This research shows that image omics combined with clinical pathology model(nomograph)has good prediction efficiency of 2D radiomics nomograph forecast 3D has the best efficiency of 2D + 3D joint model from 2D images radiomics models predict performance better,as a kind of noninvasive detection of EGFR mutation status tool,no doubt for clinical prediction EGFR mutation state provides a better choice.