Computed Tomography–Pathology Deep Learning Model for the Precise Prediction of Recurrence in Pathological Stage IA Lung Adenocarcinoma

Background: Postoperative prognosis of pathological stage IA lung adenocarcinoma (LUAD) is heterogeneous. The ability of the tumor–node–metastasis staging system to predict recurrence is limited. This study aims to develop a precise deep learning model to predict the prognosis of stage IA LUAD. Methods: The data of a consecutive cohort of patients with pathological stage IA LUAD who underwent surgical treatment at Zhongshan Hospital Fudan University were retrospectively analyzed. We propose a new deep learning model, ResNet 3D-Pathology Fusion (Res3D-PF), which is based on a three-dimensional ResNet backbone and an image–pathology fusion module, to predict recurrence-free survival (RFS) based on preoperative computed tomography (CT) images and the International Association for the Study of Lung Cancer (IASLC) grade. The predictive performance of the models was assessed by receiver operating characteristic (ROC) curve analysis. Independent predictors of RFS were identified by multivariable Cox regression analysis. Results: Overall, 551 patients with stage IA LUAD (median age, 61 years; 339 women) were included in the analysis, including 368 in the training set and 183 in the validation set. The CT-pathology deep learning model (Res3D-PF) achieved an area under the curve (AUC) of 0.837 in the validation set, significantly higher than the 8th T-stage system (AUC = 0.660) and IASLC grade (AUC 0.684). Multivariable analysis suggested that the deep learning model outputs were independent prognostic factors for RFS (hazard ratio 20.142, 95% confidence interval 3.716–109.185; p < 0.001). Patients with a high recurrence risk predicted by the model (high-risk group) were significantly associated with shorter RFS. The 5-year RFS rate of the high-risk group was significantly lower than that of the low-risk group (73.1% vs. 98.5%, p < 0.001). Conclusions: We developed and validated a deep learning model for predicting postoperative recurrence in patients with pathological stage IA LUAD based on CT images and pathological features. The model had better predictive performance than the 8th T-stage system and IASLC grade alone, making it suitable for individualized treatment strategies selection.