This study presents an ensemble deep learning model to predict lymphovascular invasion (LVI) in gastric cancer using whole-slide images (WSI) stained with hematoxylin and eosin (H&E). The model combines a classification model (ConViT) and a detection model (YOLOX) to improve the accuracy of LVI detection. The ConViT model demonstrated superior performance in both AUROC and AUPRC metrics compared to other classification models. The YOLOX model outperformed the anchor-based YOLO v3 model in detection metrics. The ensemble model, which averaged the patch-level confidence scores of the classification and detection models, achieved the best AUROC, AUPRC, and F1 scores of 0.9880, 0.9769, and 0.9280, respectively. External validation using a publicly available dataset confirmed the model's effectiveness. The study highlights the potential of deep learning in digital pathology to improve the accuracy and efficiency of LVI detection, which is crucial for better patient outcomes in gastric cancer.This study presents an ensemble deep learning model to predict lymphovascular invasion (LVI) in gastric cancer using whole-slide images (WSI) stained with hematoxylin and eosin (H&E). The model combines a classification model (ConViT) and a detection model (YOLOX) to improve the accuracy of LVI detection. The ConViT model demonstrated superior performance in both AUROC and AUPRC metrics compared to other classification models. The YOLOX model outperformed the anchor-based YOLO v3 model in detection metrics. The ensemble model, which averaged the patch-level confidence scores of the classification and detection models, achieved the best AUROC, AUPRC, and F1 scores of 0.9880, 0.9769, and 0.9280, respectively. External validation using a publicly available dataset confirmed the model's effectiveness. The study highlights the potential of deep learning in digital pathology to improve the accuracy and efficiency of LVI detection, which is crucial for better patient outcomes in gastric cancer.