Comparative Studies Of Imaging And Pathology Of Lung Pure Ground-glass Nodules

Part 1 Differentiating minimally invasive adenocarcinoma and invasive adenocarcinoma presenting as pure ground-glass nodules using a neural network model based on CT imagesObjective: Based on CT images,a neural network model was established to improve the diagnostic efficacy to differentiate minimally invasive adenocarcinoma(MIA)and invasive adenocarcinoma(IAC)presenting as pure ground-glass nodules(p GGNs).Materials and methods: We retrospectively collected 268 cases with MIA and 201 cases with IAC presenting as p GGNs on thin-slice CT confirmed by surgery and pathology at the first affiliated Hospital of Dalian Medical University from December 2012 to December 2018.The clinical and imaging features of all patients were collected and the differences of clinical and imaging features between MIA group and IAC group were compared.The inclusion of input variables and factors based on p value less than 0.05 with univariable analysis were uesd to establish a neural network model.The Receiver operating characteristic curve(ROC)and area under the curve(AUC)was used to analyze,calculate and compare to predict the diagnostic efficiency of IAC.Results: In clinic,The average age of patients and smokers in the MIA group was lower or less frequent than that in the IAC group(P = 0.000 or 0.015).In CT imaging morphological signs,the incidence of circular in the MIA group was higher,but the incidences of lobulation,air bronchogram,vacuole and vessel convergence sign were less than in the IAC group(all P≤ 0.050).In CT quantitative parameters,the CT-LP,CT-W,diameter,volume and mass in the MIA group were significantly lower than in IAC group(all P= 0.000).The AUC in the neural network model was significantly better than that in CT quantitative parameters.In the neural network model,the AUC was 0.91,accuracy 83.55%,sensitivity 82.09% and specificity 85.45%,respectively.Conclusion: It is helpful to differentiate MIA from IAC prensenting as p GGNs using a neural network model based on CT image.Part 2 Radiomic signature based on CT imaging to distinguish lepidic growth invasive adenocarcinoma from non-lepidic growth invasive adenocarcinoma in pure ground glass nodulesPurpose: A radiomic signature based on CT imaging was developed to improve the diagnostic efficacy to identify lepidic growth invasive adenocarcinoma(LGA)and non-lepidic growth invasive adenocarcinoma(NLGA)presenting as pure ground-glass nodules(p GGNs).Materials and methods: We retrospectively collected 201 cases with invasive lung adenocarcinoma(IAC)confirmed by surgery and pathology,presenting as p GGNs on thin-slice CT at the first affiliated Hospital of Dalian Medical University from December 2012 to December 2018.According to the ratio of 7: 3,they were divided into training group and test group randomly.The correlation coefficient method and the least absolute contraction and selection operator(LASSO)method were used to determine radiomic features from pre-operative CT images and established a logistic model.For clinical baseline features,we used univariate and multivariate logistic regression analysis to establish a clinical baseline model and combined clinical baseline and radiomic features to establish a combined model.Receiver operating characteristic curve(ROC)is used to analyze and evaluate the predictive value of three models to differentiate LGA and NLGA.Results: In clinical baseline characteristics,mean CT value of the largest plane(CT-LP)and mean CT value of the whole(CT-W)in the LGA group were significantly lower than those in the NLGA group(all P < 0.050),the diameter in the LGA group was larger than that in the NLGA group(P = 0.020).Only CT-W was an independent predictor to construct the clinical baseline model.The area under the curve(AUC)of clinical baseline model used to differentiate LGA and NLGA was 0.75 in the training group and 0.74 in the test group,respectively.And the AUC of the radiomic model was 0.81 and 0.76 and the AUC of the combined model was 0.82 and 0.76,respectively.The radiomic model was basically similar to the combined model.Conclusion: A radiomic signature based on CT imaging have better diagnostic performance to distinguish LGA from NLGA in p GGNs than a clinical baseline model.