Histogram Analysis Combined With CT Morphological Characteristics To Identify Visceral Pleural Invasion In Stage ? Lung Adenocarcinoma

Objective:To construct a predictive model to predict VPI in stage I lung adenocarcinoma using CT histogram analysis combined with morphological characteristics and to evaluate its diagnostic performance.Methods:A total of 351 surgically resected pathological-N0M0 lung adenocarcinomas?3 cm in size from January 2014 to November 2017 in our institution were retrospectively analyzed.Two radiologists independently evaluated CT features,including lesion location,maximum diameter of the lesion,the minimum distance from the lesion to the pleura(DLP)in three-dimensional reconstruction image and the relation of the lesion to the adjacent pleura(RAP).We classified RAP into 5 grades:1.the lesion shows no contact with the pleura;2.a line can be seen between the lesion and the pleura without retraction of the involved pleura;3.the lesion shows broad contact with the pleura,but without typical pleura tags;4.one or more linear strands radiated from the leison to the pleural surface with pleural tags;5.the lesion shows broad touch with pleura,and with pleural tags.The precise edges of the VOI were manually segmented by two thoracic radiologists using FireVoxel software,if necessary to exclude normal vessels besides nodules as much as possible.An experienced pathologist evaluated the histopathological patterns according to the7th Edition of the TNM Classification for Lung Cancer:PL0 no pleural involvement;PL1 invasion beyond the elastic layer;PL2 invasion to the surface of visceral pleura.PL1 and PL2 were combined into the VPI(+)group,while PL0 were classified as the VPI(-)group.Continuous variables were compared by Student's t-test or the Mann-Whitney U test according to whether the data conformed to a normal distribution.Differences in categorical variables were analyzed using Pearson's chi-square test and Fisher's exact test.Variables that exhibited statistically significant differences in the model-development cohort were included in multivariate logistic regression analysis,and calculated confidence interval(CI)and odds ratio(OR).The established predictive Model A,Model B,Model A+B and the diagnostic performance in the validation cohort were analyzed using receiver operating characteristic(ROC)regression analysis.The diagnostic sensitivity,specificity,accuracy and area under the ROC curve(AUC)were calculated at a cut-off value in context of the maximum Youden index.The Akaike information criterion(AIC)in?~2analysis was used to determine the best-fit model.Results:Area(OR=1.001,95%CI:1.001-1.002,P=0.002),Mean(OR=1.004,95%CI:1.002-1.005,P<0.001),Kurtosis(OR=1.228,95%CI:1.068-1.413,P=0.004),the maximum diameter(OR=2.156,95%CI:1.538-3.022,P<0.001)and DLP(minimum distance from the lesion to the pleura)(OR=0.787,95%CI:0.707-0.875,P<0.001)were identified as independent predictors of VPI.The area under the ROC(Az value),accuracy,sensitivity and specificity of model A+B were 0.935,85.2%,80.2%and88.8%,respectively,exhibiting a significantly higher Az value than either model A or model B alone(0.799 vs 0.935,P<0.001;0.869 vs 0.935,P<0.001).The predictive performance of model A+B was also significantly improved over those of model A and model B with respect to the model fit(AIC value 75.7 vs 50.7,P<0.001;75.7 vs65.9,P<0.001).Conclusions:Histogram analysis combined with morphological characteristics exhibit a superior diagnostic performance in predicting visceral pleural invasion in stage I lung adenocarcinoma.