RESUMO
Objective: To assess the value of CT texture features in differentiating minimally invasive adenocarcinoma (MIA) and invasive adenocarcinoma (IAC) manifesting as sub-solid pulmonary nodules. Methods Totally 100 patients with pulmonary adenocarcinoma (43 MIA and 57 IAC lesions) manifesting as sub-solid pulmonary nodules confirmed by pathology underwent CT scanning. The solid presence, lesion size, shape regularity and margins of pulmonary nodules were assessed to construct a subjective finding model, while 896 texture features were extracted with in-house software. Diagnostic performance of prediction models were evaluated using ROC curve analysis. Results: The solid presence and lesion size of sub-solid pulmonary nodules manifested very good coherence in subjective finding model. The solid presence (odds ratio=8.177, 95%CI [1.142, 58.575]) was proved to be an independent predictor in the subjective model. Of 896 CT texture features, 4 independent features were identified as risk factors to build the texture based model via multivariate analysis. Compared with the subjective model, the texture based model achieved better discrimination accuracy in the training set, the sensitivity, specificity and AUC of texture based model in differentiating MIA and IAC was 0.85 (33/39), 0.90 (28/31), 0.94 (95%CI [0.88,0.99]), respectively, while was 0.89 (16/18), 1.00 (12/12) and 0.97 (95%CI [0.92,1.00]) in validation set, respectively. Conclusion: CT texture based model has potential to preoperatively differentiate MIA and IAC in patients with sub-solid pulmonary nodules.