| Objective: To explore the value of differential diagnosis of lung adenocarcinoma with ground glass density nodules between HRCT signs and texture analysis parameters of thin layer reconstruction,establish prediction models of CT signs,texture analysis,CT signs and texture analysis,and evaluate the ability of identification of each model.Materials and methods: We retrospectively collected 129 cases of lung adenocarcinoma(diameter ≦ 3cm)with ground-glass density nodules on HRCT diagnosed by biopsy or surgery and histopathologically confirmed in the hospital of gejishan from December 2016 to September 2019.Lesions were divided into two groups according to the presence or absence of infiltration.Age,sex and CT findings were recorded in detail.CT signs including the occurrence of lesion location,size and CT value,shape(round or class round)and the edge of the lesion,if there are points Ye Zheng,burr,indications of the presence of pleural sag)and internal(cavity)or of the presence of air-filled bronchi,blood vessels,cluster sign or package being in the presence of),complete the processing of multidimensional reconstruction,the different forms of the edge of the lesion and inherent characteristics were analyzed.The areas of interest were delineated layer by layer on the CT image,and the texture parameters of the lesion were automatically extracted by computer software,including histogram,morphology,grayscale co-occurrence matrix,hataric,and range matrix parameters.All extracted texture parameters were reduced by LASSO analysis method to obtain texture features with differences.Univariate analysis was carried out for CT signs and texture parameters after dimension reduction.According to the kolmogorov-smirnov normal test results,two independent samples t-test and chi-square test were used for variables conforming to normal distribution,while mann-whitney U test was used for variables not conforming to normal distribution.Multivariate Logistic regression was used to establish prediction models for CT signs,texture analysis,CT signs and texture analysis,and hosmer-lemeshow goodness of fit test was used to evaluate the calibration of the prediction model.The ROC curve was drawn to evaluate the predictive performance of the model.Results:There was no significant difference in gender and age between the two groups,and both non-invasive and invasive lesions were more common in middle-aged and elderly women.Among the CT signs with statistical differences,the size,average CT value and the size of solid components in m GGN in the infiltrating group were higher than those in the non-infiltrating group(14.20±6.08 vs 9.94±4.70,-541.52±82.90 Hu vs-635.22 ±104.33 Hu,6.70±2.82 vs 4.70±1.33).The incidence of leaf segmentation and pleural depression was higher than that of the pre-invasion group(52.44% vs 29.79%,43.90% vs 17.02%).The CT sign prediction model was composed of lesion size,mean CT value and pleural depression sign.The obtained differential diagnosis model was 5.149+0.120× lesion size +0.011× mean CT value +1.321× pleural depression sign,AUC value was 0.840,sensitivity was 75.6%,specificity was 82.9%,+LR was 4.44,-LR was 0.29.1048 texture feature parameters were extracted by computer software,and 8 of the most distinctive texture parameters were obtained after LASSO dimension reduction.Finally,2 parameters were incorporated into the texture analysis model,and the obtained texture analysis prediction model was:-1.841-kurtosis*1.276+ GLCMEntropy_angle90_offset6*0.440,AUC value was 0.835,sensitivity was 73.17%,specificity was 85.11%,+LR was 4.91,-LR was 0.32.The AUC value of CT sign prediction model and texture analysis prediction model is lower than that of CT sign plus texture analysis prediction model.The AUC value is 0.900, the sensitivity is 86.59%,the specificity is 85.11%,+ LR is 5.81,-LR is 0.16.Conclusion: 1.Among HRCT morphological signs,the size of the lesion,the average CT value,and the CT sign prediction model established by the pleural depression sign have some value in identifying the lesion invasion.2.In CT texture analysis,the texture analysis prediction model established by kurtosis and GLCMEntropy_angle90_offset6 can better distinguish the infiltrability of diagnosis lesions.3.Combined with the prediction model of CT signs and texture analysis,it has a high diagnostic efficiency and sensitivity,and can accurately evaluate the invasive nature of ground glass nodular lung adenocarcinoma,so as to provide reference for further targeted diagnosis and treatment. |
PDF Full Text Request