Related Experiment Video
Updated: Oct 3, 2025
03:05
Author Spotlight: Advancing Early Detection and Treatment of Gastrointestinal Tumors
Published on: February 16, 2024
1.2K
Interpretable tumor differentiation grade and microsatellite instability recognition in gastric cancer using deep
Feng Su1, Jianmin Li2, Xinya Zhao3
1Peking-Tsinghua Center for Life Sciences, Academy for Advanced Interdisciplinary Studies, Peking University, 100871, Beijing, China.
Laboratory Investigation; a Journal of Technical Methods and Pathology
|February 18, 2022
Summary
A deep learning system accurately identifies gastric cancer subtypes and microsatellite instability (MSI) directly from standard tissue slides. This AI tool enhances diagnostic speed and interpretability, aiding in personalized cancer care.
Area of Science:
- Oncology
- Computational Pathology
- Artificial Intelligence in Medicine
Background:
- Gastric cancer exhibits significant heterogeneity, complicating accurate and timely diagnosis.
- Current diagnostic methods rely on subjective pathologist assessment or lengthy molecular tests.
- Need for efficient, objective, and interpretable diagnostic tools for gastric cancer subtypes and molecular markers.
Purpose of the Study:
- To develop and validate a deep learning (DL) system for classifying gastric cancer differentiation grade and recognizing microsatellite instability (MSI).
- To utilize hematoxylin-eosin (HE) stained whole-slide images (WSIs) for direct, interpretable tumor analysis.
- To assess the system's accuracy and interpretability in diverse patient cohorts.
Main Methods:
- A DL system was trained on 348 annotated WSIs from 467 gastric cancer patients.
- Tumor differentiation (poorly vs. well-differentiated adenocarcinoma) was assessed based on glandular structure features.
- Microsatellite instability (MSI) status was recognized directly from HE-stained WSIs in independent testing cohorts.
Main Results:
- DL models achieved high F1 scores for differentiating tumor grades (0.8615 for poorly differentiated, 0.8977 for well-differentiated).
- The system demonstrated interpretability by focusing on glandular structure formation.
- Patient-level accuracy for MSI recognition directly from WSIs was 86.36% in the testing cohort and 83.87% in an integrated cohort.
Conclusions:
- The developed DL system accurately classifies gastric cancer differentiation grade and MSI status using HE-stained WSIs.
- The system offers enhanced interpretability, crucial for clinical adoption.
- This AI-powered approach shows potential for improving the efficiency and accuracy of gastric cancer diagnostics in clinical practice.
