Artificial intelligence-based reclassification of gastric adenocarcinoma enables prognostic stratification via...

Area of Science:

• Oncology
• Computational Pathology
• Artificial Intelligence

Background:

• Gastric adenocarcinoma (GAC) presents significant diagnostic and prognostic challenges due to its heterogeneity.
• The traditional Laurén classification for GAC has limitations, including interobserver variability, especially for mixed subtypes.
• Artificial intelligence (AI) offers a potential solution for standardizing GAC diagnosis and improving prognostic accuracy.

Purpose of the Study:

• To develop and validate an AI-driven framework for automated classification and prognostic stratification of GAC.
• To assess the utility of AI-based image analysis in overcoming limitations of manual GAC subtyping.
• To establish a novel, reproducible prognostic marker for GAC patients.

Main Methods:

• Retrospective analysis of 404 resected GAC cases (2015-2022) using whole-slide images (WSIs).
• Training a two-stage deep learning pipeline (YOLO26m-cls) to distinguish malignant from non-malignant tissue and classify malignant patches as intestinal or diffuse.
• Development of the diffuse prognostic score (DPS) based on the proportion of diffuse-type patches and correlation with overall survival (OS).

Main Results:

• The AI models achieved high diagnostic performance: GAC-I accuracy 0.9437 ± 0.0317, GAC-ST F1 score 0.7528 ± 0.1094.
• A diffuse prognostic score (DPS) ≥ 0.5 was significantly associated with lower median overall survival (16.1 months vs. 42.067 months).
• The DPS remained an independent predictor of mortality after adjustment (HR 2.684, p=0.027), confirmed by multivariate Cox-regression and case-control analyses.

Conclusions:

• A robust AI framework was developed and validated for automated GAC classification and prognostic stratification using H&E WSIs.
• The diffuse prognostic score (DPS) is an independent and reproducible marker of overall survival in GAC.
• The DPS holds potential for integration into clinical pathology workflows to guide personalized GAC treatment strategies.