Application of artificial intelligence based on contrast-enhanced CT imaging for predicting peritoneal metastasis in patients with T3/T4 stage gastric cancer.

Abstract

Gastric cancer, prevalent in East Asia, often presents with peritoneal metastasis at diagnosis, limiting surgical options and reducing survival rates. Given the low sensitivity of current diagnostic methods, this study aimed to develop and evaluate deep learning models based on preoperative contrast-enhanced computed tomography images to improve the detection of occult peritoneal metastasis in T3/T4 stage gastric cancer. We first evaluated the performance of several convolutional neural network architectures and identified Inception-ResNetV2 as the best-performing model. To further optimize the model's performance, we integrated multiple attention mechanism modules, with the SE module showing the most significant improvement. The SE-augmented Inception-ResNetV2 model achieved a receiver operating characteristic area under the curve of 0.973, Precision-Recall area under the curve of 0.908, and an F1-Score of 0.818, outperforming all other models. Calibration curves demonstrated good agreement between predicted and actual outcomes, while decision curve analysis highlighted the model's clinical utility. These findings suggest a potential approach for improving clinical predictive modeling by integrating advanced deep learning architectures with attention mechanisms. For patients identified as high-risk, further staging laparoscopy is recommended to minimize unnecessary surgery and guide treatment decisions.