Related Experiment Video
Updated: Aug 6, 2026
10:31
Intraoperative Gastroscopy for Tumor Localization in Laparoscopic Surgery for Gastric Adenocarcinoma
Published on: August 9, 2016
Multiscale Spatial Fusion Feature-Driven Characterization of Gastric Cancer Invasive Margins: A Multicenter Cohort
Guoliang Zheng1, Xiaomiao Chai2, Peng Jin3
1Department of Gastric Surgery, Cancer Hospital of Dalian University of Technology, Liaoning Cancer Hospital & Institute, Shenyang, Liaoning, China.
Advanced Science (Weinheim, Baden-Wurttemberg, Germany)
|July 20, 2026
Summary
A new AI model, GAVR, accurately differentiates advanced gastric cancer stages (T4a/b) using multi-scale spatial features. This improves surgical planning and prognosis by enhancing radiologist accuracy and reducing reading time.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Accurate preoperative staging of gastric cancer T4a/b is critical for treatment planning and patient outcomes.
- Conventional CT imaging has limitations in assessing peritumoral microinvasion, impacting staging accuracy.
Purpose of the Study:
- To develop and validate a novel multi-scale spatial feature fusion model (GAVR) for precise preoperative differentiation of gastric cancer T4a/b stages.
- To assess the clinical utility and generalizability of the GAVR model in a multicenter cohort.
Main Methods:
- Proposed the GAVR model with a three-tier architecture: Boundary-Augmented U-Net for extended regions of interest (eROI) generation, parallel pathways for radiomics and deep learning features (2D/3D), and a Vision Transformer for feature fusion.
- Validated the model on a multicenter cohort of 1804 patients (internal, external, prospective sets) and conducted a blinded reader study with 16 radiologists.
Main Results:
- GAVR achieved high generalizability with AUCs of 0.987 (external) and 0.987 (prospective).
- Radiologist diagnostic accuracy significantly improved from 0.609 to 0.795 with GAVR assistance, and reading time decreased by 60%.
- Ablation studies confirmed the importance of multi-scale features for characterizing invasive margins and mitigating overfitting.
Conclusions:
- The GAVR model effectively fuses multi-scale spatial features to characterize gastric cancer invasive margins, offering precise T4a/b staging.
- GAVR demonstrates significant translational value as an embeddable decision-support tool for multidisciplinary gastric cancer management.