Clinical Study On Differentiation Of Infiltration Degree Of Lung Adenocarcinoma With Ground Glass Nodules Based On CT Imaging

Objective Different pathological types of lung adenocarcinoma have different treatment plans.Accurate identification of the degree of invasion of lung adenocarcinoma before operation is beneficial to individual accurate diagnosis and treatment,but the overlapping imaging findings of ground glass nodular(GGN)lung adenocarcinoma are difficult to distinguish clinically.Therefore,based on the extraction of imaging features from computed tomography(CT)images,the application value of imaging in predicting early lung adenocarcinoma with GGN was discussed.The purpose of this study is to combine imaging science with clinical information and CT image features to construct the best prediction model and to establish an individual diagnostic line chart.Methods A total of 180 GGN in 167 patients with primary lung adenocarcinoma with surgical and pathological results in the affiliated Hospital of Inner Mongolia Medical University from October 2016 to February 2022 were analyzed retrospectively,including atypical adenomatoid hyperplasia(AAH),adenocarcinoma in situ(AIS),microinvasive adenocarcinoma(MIA)and invasive adenocarcinoma(IA).The first two types were classified as proglandular lesions(PGL).First,the IA was distinguished by the diagnosis of non-IA group and IA group,then the classification of PGL group and MIA group was carried out to distinguish them,and finally the differential diagnosis of three pathological types of lung adenocarcinoma was realized.Univariate analysis and multivariate logistic regression were used to screen independent predictors for clinical information and CT image features of all data.The focus part of the CT image was segmented manually,and the imaging features of the lesions were extracted from the volume of the region of interest by Pyradiomics software,and then the optimal set of features was obtained by univariate analysis,Spearman correlation analysis,minimum absolute contraction selection operator and multi-factor stepwise regression.The clinical imaging model,imaging model and joint model were established by multi-factor logical regression,and the areas under the subjects’ working characteristic curves of the three models were compared.the clinical benefits were determined by comparing the predictive performance of the calibration curve and the decision curve.Then multi-factor stepwise regression was used to construct the individual diagnostic line map of lung adenocarcinoma.Results Non-IA group and IA group: after screening by univariate analysis and multi-factor logical regression,the maximum diameter,average CT value and pleural traction sign of GGN were independent predictors of IA.The AUC of the clinical imaging model constructed based on this was 0.795(95% CIM 0.713-0.878)and 0.815(95% CI 0.696-0.934)in the training set and verification set,respectively.A total of 1595 imaging features were extracted from the lesions,and the best subset containing five features was obtained after step-by-step screening.these features were weighted according to a certain coefficient to obtain imaging scores and build an imaging model.The AUC in the training set and verification set were 0.854(95% CIJ0.784-0.925)and 0.830(95% CILI 0.718-0.942),respectively.A joint model is constructed by combining CT image features and image group scores.The AUC training set is 0.874(95%CIPower0.809-0.938)and the verification set is 0.848(95% CIPower0.738-0.959).In the PGL group and the MIA group,the average CT value of GGN screened by the same method can distinguish PGL from MIA.The AUC of the clinical image model constructed according to it in the training set and verification set is 0.619(95% CIMagi 0.445-0.793)and 0.670(95%CIBE 0.425-0.915),respectively.The 1595 imaging features extracted from the lesions were screened to get 5 best features,and the features were weighted to get the imaging score.The AUC of the constructed imaging model in the training set and verification set was 0.845(95%CILI 0.732-0.957)and 0.791(95% CILI 0.550-1.000),respectively.A joint model is constructed by combining CT image features and image group scores.The AUC training set is0.829(95% CIPower0.706-0.951)and the verification set is 0.890(95%CIPower0.745-1.000).The calibration curve shows that the prediction performance of the two combined models is the best.The decision curve shows that the clinical benefits of the two groups of imaging models and combined models are similar,and are significantly better than the clinical imaging models.Conclusion(1)the maximum diameter,average CT value and pleural traction sign of GGN are independent predictors of IA.The average CT value of GGN can be used as a quantitative index to predict the three pathological types of lung adenocarcinoma.(2)the imaging model can be used to predict the pathological type of GGN before operation,and it has good diagnostic performance.In addition,the image features filtered by the algorithm are more helpful to the establishment of the ensemble model than the original image features.(3)the diagnostic efficiency of the joint model constructed by the imaging group model combined with CT image features can be further improved.As a non-invasive and accurate prediction tool,the line chart shows a wide range of application value,and is expected to guide clinicians to choose the best intervention measures in the individual evaluation and management of GGN.