Abstract
INTRODUCTION: Accurate prediction of the lymphovascular invasion (LVI) status in patients with T1-stage invasive lung adenocarcinoma (LUAD) is crucial for treatment decision-making. Currently, there is a lack of highly efficient and precise prediction models. METHODS: In this retrospective study, 334 patients with T1-stage invasive LUAD who underwent radical surgery from four academic medical centers were included. Conventional radiomic features, two-dimensional deep learning (2D DL) features, and three-dimensional deep learning (3D DL) features were extracted from the tumor regions of the patients' CT images. Corresponding prediction models were constructed, and these features were integrated to develop a combined model for identifying the LVI status. The performance of the model was evaluated by calculating the area under the receiver operating characteristic (ROC) curve (AUC), and the net benefit of the models was compared using decision curve analysis (DCA). RESULTS: The combined model demonstrated excellent performance in distinguishing the LVI status, with its predictive ability superior to that of individual models. The AUC values for the training set, internal validation set, and external test set reached 0.958 (95% CI: 0.9294 - 0.9863), 0.886 (95% CI: 0.7938 - 0.9786), and 0.884 (95% CI: 0.8277 - 0.9401), respectively. DCA showed that the net benefit provided by the combined model was higher than that of other radiomic models. CONCLUSIONS: The combined model integrating radiomics, 2D DL, and 3D DL exhibits excellent performance in predicting the LVI status of patients with T1-stage invasive LUAD, and can provide key information for clinical treatment decision-making.