This study shows that deep learning can predict leakage during laparoscopic gastrectomy for gastric cancer, improving surgical outcomes.

• Leakage rates were 1.3% at the duodenal stump and 4.3% at the esophagojejunal anastomoses.
• The ResNet18 model achieved a recall of 84.7% for stump leakage and 80% for anastomosis leakage when trained on high-quality, single-image datasets with augmentation.

Using AI during surgery could facilitate timely interventions, enhancing patient safety and outcomes.

• Key image features, both local and surrounding tissues, were critical for accurate predictions.