Application Value Of Prediction The Invasiveness Of Lung Adenocarcinoma Presenting As Ground-glass Nodules By CT Diagnosis And Radiomics Analysis

Part one The value of CT signs in evaluating the invasive risk factors of lung adenocarcinoma presenting as ground-glass nodulesObjective:Through the analysis of ground-glass nodules(GGN)multi-phase scanning CT signs and their relationship with tumor invasiveness,explore the risk factors of lung adenocarcinoma presented as GGN and the diagnostic value of CT signs on early lung adenocarcinoma.Methods:A retrospective analysis of patients with GGN performed by CT of the lung from May 2017 to March 2020 in the Second Affiliated Hospital of Suzhou University.All the clinical,pathological,and imaging data of patients were collected,while the preoperative CT appearance showed as GGN and postoperative pathological results confirmed as lung adenocarcinoma or atypical adenomatous hyperplasia(AAH).All the patients underwent plain scan and dual-phase enhanced CT scan within one month before the operation.161 cases of GGN in total were enrolled and divided into two groups according to whether they had invasive component in the postoperative pathological results.There were 7 cases of AAH and 33 cases of adenocarcinoma in situ(AIS)in the pre-invasive group,while the invasive group included 71 cases of minimally invasive adenocarcinoma(MIA)and 50 cases of invasive adenocarcinoma(IAC).The position,shape,nature,imaging signs and average diameter of the target GGN were observed and recorded in the CT plain scan images.The average CT value of GGN and lung tissue on the same plane were measured in plain scan and dual-phase enhanced CT images.The relative CT value of GGN in each phase,the net increase of CT value and relative CT value of GGN in arterial phases and venous phases were getted by calculation.Finally,some statistical methods such as the receiver operating characteristic(ROC)and logistic regression analysis were used to analyze the difference between the two groups,to explore the independent risk factors of the invasion of GGN lung adenocarcinoma,to evaluate the diagnostic value of the CT signs and the changes in the CT values of each stage between the pre-invasive and invasive lung adenocarcinoma.Results:Between the pre-invasive group and invasive group,the nature(P=0.003),shape(P=0.019),lobular sign(P=0.005),bronchial air sign(P=0.002),pleural depression sign(P=0.015),abnormal vessel sign(P=0.000),average diameter(P=0.000),plain scan CT value(P=0.004)and relative CT value(P=0.004),arterial phase CT value(P=0.003)and relative CT value(P=0.001),the venous phase CT value(P=0.006)and the relative CT value(P=0.002)all had statistically significance.The ROC curves of continuous variables were plotted and the cut-off values were obtained through the Youden index.The area under curve(AUC)of the average GGN diameter was 0.748,the cut-off value was 9.37mm,the sensitivity was 0.653,and the specificity was 0.825.The AUC of CT value and relative CT value of each stage were less than 0.7,which showed that the diagnostic efficiency was poor.Multivariate analysis indicated that average diameter(P=0.003),bronchial air sign(P=0.048)and abnormal vessel sign(P=0.035)were independent invasive risk factors,the OR values were 6.592,3.148 and 9.867,respectively.Conclusion:CT signs were helpful to distinguish whether GGN lung adenocarcinoma had invasiveness or not.The average diameter,bronchial air signs,and abnormal vessel signs were independent risk factors for the invasion of GGN lung adenocarcinoma.Part two The value of radiomics models in predicting the invasiveness of lung adenocarcinoma presenting as ground-glass nodulesObjective:To explore the diagnostic efficacy and clinical application feasibility of radiomic models based on multi-phase scanning CT images in predicting the invasiveness of lung adenocarcinoma presenting as GGN.Methods:The enrolled cases and grouping situation were the same as the first part.The non-contrast phase(NCP),arterial phase(AP)and venous phase(VP)CT images of all the 161 GGN patients were retrospectively collected.ITK-SNAP software was used to manually segment the region of interest(ROI)area of all the CT images in DICOM format,then the Pyradiomics toolkit was used to automatically extract radiomics features of all the lesions.Three single-phase models(NCP model,AP model and VP model)based on three single-phase CT images and a combined model based on multi-phase CT images were established by the highest correlation radiomics features selected through the least absolute shrinkage and selection operator(LASSO)approach.The logistic regression classifier was then used to train and verify all the models by 5-fold cross-validation.The performances of these models were evaluated through the ROC analysis,two-tailed t test and decision curve analysis(DCA).Results:There were 1083,978 and 1119 radiomics features respectively extracted from the plain scan,arterial phase and venous phase of 161 cases.Through LASSO analysis,4,5 and 4 features were selected and used for the development of radiomics models from NCP,AP and VP CT images,respectively.All the radiomics features in the NCP model were texture features,the radiomics features in the AP model included 1 morphological feature and 4 texture features,and the radiomics features in the VP model included 1 first-order feature and 3 texture features.The combined model included all the radiomics features selected from the plain scan and dual-phase enhanced images.After ROC analysis,the AUC values of the NCP,AP,VP model and the combined model in the training data set were 0.775,0.809,0.805 and 0.840,and the AUC values in the validation data set were 0.664,0.754,0.741 and 0.794,respectively.The combined model showed better performance than NCP,AP and VP models.Two-tailed t test was performed on the combined model paired with three single-phase models,the results showed that the difference between the combined model and the NCP model was statistically significant.The analysis of the decision curve showed that when the threshold probability was greater than 60%,all the four groups of models could obtain greater net benefits in predicting the invasiveness of GGN.when the threshold probability was greater than 75%,the net income of combined model was better than the other three single-phase models.Conclusion:The NCP model,AP model and VP model all had excellent diagnostic performance in predicting the invasiveness of GGN.Among them,the diagnostic performance of AP model and VP model were better than the NCP model.Compared with three single-phase models,the combined model based on three-phase CT images showed favorable diagnostic performance and had potential to be used as a promising non-invasive tool for predicting the invasiveness of GGN.