Accurate prediction of lymph node metastasis (LNM) in intrahepatic cholangiocarcinoma (ICC) is crucial for both treatment planning and prognosis assessment. Given that visual 18F-FDG PET/CT interpretation has inherent limitations, we developed a multimodal model that integrates 3D PET/CT data with clinical patient information. To ensure the model focuses on relevant physiological features, a ‘body masking’ preprocessing pipeline was implemented to eliminate background noise and scanning bed artifacts. Our model achieved a 0.80 AUROC and 0.78 accuracy, significantly outperforming the diagnostic accuracy of clinicians (0.61). This study suggests that deep learning models can serve as reliable computer-aided diagnostic tools to assist clinicians’ decision-making in clinical settings.