| Objective Preoperative accurate assessment of invasive lung adenocarcinoma with ground glass nodules is of great significance for clinical decision and prognosis of patients.However,there are still great challenges in the field of clinical imaging diagnosis.Therefore,we intended to extract radiomics characteristics from computed tomography images to distinguish lung adenocarcinoma invasion as ground-glass nodules.Furthermore,the best diagnostic model was established by combining the prominent radiomics features with the traditional clinical features to provide clinical guidance for ground-glass pulmonary nodules.Methods This retrospective study researched 93 patients with 103 ground glass pulmonary nodules,which were surgically confirmed at the Affiliated Hospital of Inner Mongolia Medical University from October 2013 to January 2021.The ’gold standard’ for the diagnosis of all patients was based on the pathologic findings of the final paraffin section,including atypical adenomatous hyperplasia,adenocarcinoma in situ,minimally invasive adenocarcinomas,and invasive adenocarcinoma,with the first three classified as preinvasive lesions and the latter as invasive lesions.The routine clinical risk factors and thin-slice CT imaging characteristics were screened by univariate and multivariate analysis successively.Three-dimensional ground glass nodules were manually segmented from thin-slice CT images.The open source software pyradiomics was used to automatically read the lesions and extract the radiomics features,which were screened by univariate analysis,correlation analysis and multivariate analysis.In the end,logistic regression analysis was used to establish three diagnostic models,namely clinical imaging basic model,radiomics model and combination model,and the diagnostic efficacy was assessed according to the area under curve of the receiver operating characteristic curve.Results Univariate analysis of 5 routine clinical risk factors and 12 CT imaging features showed that the maximum diameter,mean CT attenuation value,solid component,irregular shape,burr sign,lobulation sign,pleural pull sign,air bronchial sign,vascular crossing sign and vascular cluster sign could indicate the invasion of GGN.After multivariate analysis,the independent predictors of invasive adenocarcinoma were the maximum diameter and mean CT attenuation value.The conventional model of clinical imaging established by this method had good diagnostic performance,with AUC of 0.838(training group)and 0.810(validation group).log.sigma.5.0.mm.3D＿glszm＿Low Gray Level Zone Emphasis and Wavelet.HLL＿Firstorder＿Kurtosis are two quantitative radiomics features with significant correlation to invasive adenocarcinoma(P<0.05),and the AUC of the radiomics model was 0.869(training group)and 0.847(validation group).The combined model based on significant clinical imaging features and radiomics scores was significantly better than the clinical imaging model and the radiomics model in differentiating invasive adenocarcinoma from non-invasive adenocarcinoma,with an AUC of 0.886(training group)and 0.852(validation group).Analysis of calibration curves and decision curves shows the potential benefits of using this model in clinical diagnosis.Conclusion(1)The maximum diameter and mean CT attenuation value in GGN were significantly correlated with the infiltration of pulmonary ground glass nodules.(2)Radiomics texture features extracted from CT images can be used to predict and evaluate whether GGN is invasive before surgery,which is a relatively accurate and reliable method.(3)The combined model established by radiomics scores and important clinical factors was helpful for the preoperative differentiation of invasive GGN lesions and thus became an important reference for clinical decision. |
