Computerized texture analysis predicts histological invasiveness within lung adenocarcinoma manifesting as pure ground-glass nodules

Background Differentiating histological invasiveness of lung adenocarcinoma that manifest as pure ground-glass nodules (pGGNs) is challenging. Purpose To investigate the value of computerized texture analysis for predicting histological invasiveness of pulmonary adenocarcinoma that manifest as pGGNs. Material and Methods The study consisted of 138 patients with 142 pathologically confirmed lung adenocarcinomas who had undergone computed tomography (CT) imaging. Each nodule was manually segmented and 96 texture features were extracted automatically. Hierarchical cluster analysis, the ReliefF method, and a logistic regression model were used for dimension reduction and feature selection. Performance of the texture features was evaluated by receiver operating characteristic (ROC) curve analysis. Results Pathologic analysis confirmed 26 adenocarcinomas in situ (AISs), 71 minimally invasive adenocarcinomas (MIAs), and 45 invasive adenocarcinomas (IACs). Seven best features (10 percentile, maximum 3D diameter, surface volume ratio, elongation, maximum probability, large area low gray level emphasis, and zone entropy) were chosen by using hierarchical cluster analysis and the ReliefF method. Multivariate logistic regression analysis revealed larger maximum 3D diameter, lower surface volume ratio, and higher zone entropy as independent differentiators of IACs (adjusted odds ratio [OR] = 1.59, P = 0.011; OR = 0.47, P = 0.002; OR = 6.78, P = 0.001, respectively). The accuracy based on the logistic regression model using these features for differentiating IAC from AIS/MIA reached 78.7% with ROC analysis (AUC = 0.861; sensitivity = 78.0%; specificity = 80.0%). Conclusion In patients with pGGN, computerized texture analysis has the potential to differentiate histological invasiveness; maximum 3D diameter, surface volume ratio, and zone entropy in particular are independent differentiators of IACs from AISs and MIAs.