An ensemble deep learning model was developed to predict lymphovascular invasion (LVI) in gastric cancer using H&E-stained whole-slide images. The model combines classification and detection sub-models, with the classification model (ConViT) achieving the highest AUROC and AUPRC. The detection model (YOLOX) showed slightly lower performance than ConViT when using augmented patch-level confidence scores. The ensemble model, which combines predictions from both sub-models, demonstrated the best AUROC, AUPRC, and F1 scores (0.9880, 0.9769, and 0.9280, respectively). The model was validated on an external dataset, showing improved performance compared to classification and detection models. The ensemble model outperformed single models in predicting LVI, with AUROC improvements of 2.8% and 5.9% compared to classification and detection models, respectively. The model's ability to detect LVI foci in whole-slide images has potential applications in precision medicine, improving diagnostic accuracy and reducing inter-observer variability. The study highlights the effectiveness of deep learning in histopathological analysis and its potential to enhance the diagnosis of gastric cancer.An ensemble deep learning model was developed to predict lymphovascular invasion (LVI) in gastric cancer using H&E-stained whole-slide images. The model combines classification and detection sub-models, with the classification model (ConViT) achieving the highest AUROC and AUPRC. The detection model (YOLOX) showed slightly lower performance than ConViT when using augmented patch-level confidence scores. The ensemble model, which combines predictions from both sub-models, demonstrated the best AUROC, AUPRC, and F1 scores (0.9880, 0.9769, and 0.9280, respectively). The model was validated on an external dataset, showing improved performance compared to classification and detection models. The ensemble model outperformed single models in predicting LVI, with AUROC improvements of 2.8% and 5.9% compared to classification and detection models, respectively. The model's ability to detect LVI foci in whole-slide images has potential applications in precision medicine, improving diagnostic accuracy and reducing inter-observer variability. The study highlights the effectiveness of deep learning in histopathological analysis and its potential to enhance the diagnosis of gastric cancer.