Prediction of visceral pleural invasion of clinical stage IA lung adenocarcinoma based on computed tomography features.

Background: In lung cancer, preoperative prediction of visceral pleural invasion (VPI) is helpful for choosing the best treatment plan and improving the prognosis of patients. This study aimed to investigate the usefulness of computed tomography (CT) features in predicting VPI in clinical stage IA peripheral lung adenocarcinoma (LUAD) with pleural contact. Methods: This study divided the type of contact between tumor and pleura into indirect and direct contacts. This study retrospectively analyzed patients with clinical stage IA peripheral LUAD in three hospitals and enrolled 485 patients. The CT features of lesions were analyzed to predict VPI, including relative pleural features, tumor signs, and characteristics between the tumor and pleura. Univariate and multivariate logistic regression analyses were used to select the best combination of variables to predict VPI, and the prediction models were developed. Results: The multivariate logistic regression analysis identified solid component size, pleural tag type, and vascular convergence sign to be independent risk factors for VPI in indirect pleural contact type. The area under curve (AUC) values of the model for predicting VPI in the training, internal validation, and external validation sets were 0.887, 0.799, and 0.862, respectively. Solid component size and pleural indentation sign were identified as independent risk factors for predicting VPI in direct pleural contact type. The AUC values of the model for predicting VPI in the training, internal validation, and external validation sets were 0.903, 0.848, and 0.842, respectively. Conclusions: CT predictors associated with VPI differ based on the type of contact with the pleura. The multivariate logistic regression models utilizing CT features demonstrates acceptable diagnostic accuracy in predicting VPI in clinical stage IA LUAD with pleural contact.