Overview

Gastric cancer is a leading cause of cancer-related mortality, and radical surgery remains the primary treatment. However, postoperative complications are common and can significantly impact patient recovery and quality of life. Currently, doctors lack precise tools to accurately predict which patients are at high risk for developing severe complications before surgery.

This study aims to validate a novel artificial intelligence (AI) model called "DeepComp." The DeepComp model integrates clinical data with advanced radiomic features derived from routine preoperative CT scans. Specifically, it analyzes both the tumor characteristics and the patient's body composition (including skeletal muscle and fat distribution) to assess physiological reserve.

In this prospective, multicenter observational study, researchers will enroll patients scheduled for gastric cancer surgery across five medical centers. The DeepComp model will be used to predict the risk of moderate-to-severe postoperative complications (Clavien-Dindo grade II or higher). These predictions will then be compared with the actual clinical outcomes observed 30 days after surgery. The goal is to determine the accuracy and reliability of the DeepComp model in a real-world clinical setting, potentially providing a powerful tool for personalized surgical risk assessment.

Eligibility

Inclusion Criteria:

Age ≥ 18 years.

Histologically confirmed gastric adenocarcinoma.

Scheduled for elective radical gastrectomy (open, laparoscopic, or robotic) with curative intent.

Standard preoperative contrast-enhanced abdominal CT scans (venous phase) performed within 14 days prior to surgery.

Willingness to sign informed consent.

Exclusion Criteria:

Emergency surgery due to perforation, obstruction, or massive bleeding.

Intraoperative findings of distant metastasis (Stage IV) or unresectable disease preventing R0 resection.

Concurrent or previous malignant tumors within the last 5 years (except gastric cancer).

Pregnancy or lactation.

Severe metallic artifacts on CT images preventing radiomic analysis.