ResGAT: A Residual Graph Attention Network for Cancer Subtype Classification in Whole Slide Images

Zhenhan Lin, Hao Tong, Yunfei Hu, Xianyong Sean Gui, Jeanne Shen, Byrne Lee, Lu Zhang, Daniel Moyer, Mu Zhou, Xin Maizie Zhou, Konstantinos Votanopoulos

Keywords: whole slide image classification, multiple instance learning, residual graph attention framework, cross-site generalization

Abstract: Multiple instance learning (MIL) provides a weakly supervised framework for whole slide image (WSI) classification, enabling slide-level prediction from gigapixel images with only slide-level labels. However, WSI subtype classification in realistic settings is still challenging. In this work, we propose ResGAT, a residual graph attention framework that operates on hybrid $k$-NN patch graphs and models WSI representations with stacked residual graph attention blocks. ResGAT is evaluated on the subtype classification task across a rare, class-imbalanced appendiceal cancer cohort, BRACS and two TCGA datasets. It outperforms SOTA MIL baselines on the appendiceal cancer and BRACS cohorts, and remains competitive on the TCGA datasets. On the appendiceal cancer cohort, we further assess cross-site generalization via few-shot adaptation under source shift, showing that ResGAT adapts effectively to new domains with limited labels. An ablation study is provided to validate the effectiveness of key architectural components of our method.

Primary Subject Area: Application: Histopathology

Secondary Subject Area: Learning with Noisy Labels and Limited Data

Registration Requirement: Yes

Reproducibility: https://github.com/maiziezhoulab/ResGAT

Visa & Travel: Yes

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 369