What question did this study set out to answer?

The aim is to predict molecular subtypes of gastric cancer using a weakly supervised framework from whole-slide images.

April 5, 2026Cancer Research0 citations

Abstract 1445: Predicting TCGA molecular subtypes of gastric cancer from H&E whole-slide images using a weakly supervised transMIL-attention framework.

YJYesul Jeong DBDewan Muhammad BappyIncheon National University SASangjeong AhnKorea University

Key Points

The aim is to predict molecular subtypes of gastric cancer using a weakly supervised framework from whole-slide images.
Implemented attention-based multiple instance learning (ACMIL) for WSI classification
Proposed a hybrid model combining transformer-based multiple instance learning (TransMIL) with multi-branch attention (MBA)
Trained and evaluated on 484 H&E WSIs with TCGA molecular subtype labels
Assessed model performance using AUC, accuracy, confusion matrices, and UMAP visualization
TransMIL-MBA model achieved an AUC of 0.89, outperforming ACMIL's AUC of 0.87
Accuracy improved to 0.74 for TransMIL-MBA compared to ACMIL's 0.67
Enhanced subtype separability observed in UMAP embedding with V-measure of 0.74 for TransMIL-MBA versus 0.65 for ACMIL

Abstract

Abstract Background: The Cancer Genome Atlas (TCGA) has defined four molecular subtypes of gastric cancer: Epstein-Barr virus (EBV)- associated, microsatellite instability (MSI)- associated, genomically stable (GS), and chromosomal instability (CIN). These subtypes have distinct clinicopathologic and therapeutic implications. However, the routine determination of these subtypes still relies on multimodal molecular assays that are costly and not universally available. We aimed to develop a weakly supervised deep learning framework that predicts TCGA molecular subtypes directly from hematoxylin and eosin (H accuracy: 0.74 vs. 0.67). The hybrid model also showed improved subtype separability in the UMAP embedding (V-measure 0.74 with TransMIL-MBA vs. 0.65 with ACMIL) and clearer confusion matrices. Conclusions: A weakly supervised multiple instance learning framework combining TransMIL with MBA shows promising performance for predicting TCGA molecular subtypes of gastric cancer directly from H Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 1445.

AIに質問

Bookmark

View Full Paper