Abstract: Objective: Lateral lymph node metastasis (LLNM) in papillary thyroid carcinoma (PTC) significantly impacts PTC prognosis and treatment strategies. This study aimed to develop and validate a multimodal artificial intelligence (AI)-driven integrated model to predict LLNM by combining clinical data, radiomics features, and deep learning-based imaging features. Methods: A cohort of 1,566 patients was divided into training, validation and test sets. To indirectly assess LLNM risk based on tumor aggressiveness characteristics, radiomics features were extracted from ultrasound images of the primary thyroid tumor and selected using Lasso regression to construct a support vector machine (SVM) model. Clinical variables were used for logistic regression. Deep learning features characterizing the primary thyroid nodule were derived from the TresNet model, and these predictions were integrated into a multimodal model and evaluated using ROC curves. Results: The integrated model outperformed individual models, achieving the highest area under the curve in the training (0.984), validation (0.943), and test (0.951) datasets. SHapley Additive exPlanations (SHAP) analysis identified key predictive factors such as TresNet scores, SVM-based radiomics features, tumor size, location, sex, and thyroglobulin antibody levels. After incorporating the model, the diagnostic accuracy enhanced for ultrasound physicians at all levels (junior, medium, and senior), with significant improvements in sensitivity, specificity and confidence. Conclusions: The proposed AI-driven integrated model, which predicts LLNM risk by analyzing primary thyroid tumor characteristics, demonstrated robust predictive performance and clinical utility in LLNM risk stratification for PTC patients. It significantly improves diagnostic accuracy compared to individual models, providing a non-invasive, personalized tool to support clinical decision-making.