Integrated Radiologic, Pathologic, and Transcriptomic Analysis of Lymph Node Metastasis Potential in Small (≤3 cm) Lung Adenocarcinoma

Background: This study aimed to develop and validate a multimodal radiology–pathology model (MoM-LNM) that integrates preoperative computed tomography (CT) radiomics and postoperative pathomics to identify lymph node metastasis (LNM) in lung adenocarcinoma (LUAD). Methods: Patients were enrolled from two medical centers (Center A: n = 483; Center B: n = 153). Data from Center A were divided into training (n = 339), validation (n = 72) and internal test (n = 72) sets, while Center B served as the external test set. Radiomics, pathomics, and MoM-LNM were developed using a multilayer perceptron. The optimal cutoff value for risk stratification was determined from the best-performing model. Bulk RNA sequencing and single-cell RNA sequencing were used to explore the underlying biological characteristics. Results: The MoM-LNM achieved AUCs of 0.879–0.944 across the validation and test sets and showed the best overall performance. The high-risk group had significantly worse 3-year overall survival (92.1% vs. 98.8%) and recurrence-free survival (RFS) (73.8% vs. 90.6%) than the low-risk group (both p < 0.001). Among patients without LNM, the high-risk group remained independently associated with worse RFS (hazard ratio = 2.387, 95% confidence interval: 1.350–4.219). Bulk RNA sequencing and single-cell RNA sequencing showed distinct immune characteristics in the high-risk group. Conclusions: The MoM-LNM provides postoperative risk stratification in LUAD patients. Notably, among patients without pathologically confirmed LNM, it identified a subgroup at elevated risk of adverse outcomes.