GrapHist: Graph Self-Supervised Learning for Histopathology

TL;DR

GrapHist introduces a biologically-informed graph self-supervised learning framework for histopathology that captures tissue structure and heterogeneity, achieving high performance with fewer parameters and enabling diverse diagnostic tasks.

Contribution

This work presents GrapHist, a novel graph-based self-supervised learning method tailored for histopathology, integrating cell graphs and heterophilic GNNs for improved tissue modeling.

Findings

Achieves competitive performance on multiple histopathology tasks.

Outperforms fully-supervised graph models in cancer subtyping.

Requires four times fewer parameters than vision-based models.

Abstract

Self-supervised vision models have achieved notable success in digital pathology. However, their domain-agnostic transformer architectures are not originally designed to account for fundamental biological elements of histopathology images, namely cells and their complex interactions. In this work, we hypothesize that a biologically-informed modeling of tissues as cell graphs offers a more efficient representation learning. Thus, we introduce GrapHist, a novel graph-based self-supervised learning framework for histopathology, which learns generalizable and structurally-informed embeddings that enable diverse downstream tasks. GrapHist integrates masked autoencoders and heterophilic graph neural networks that are explicitly designed to capture the heterogeneity of tumor microenvironments. We pre-train GrapHist on a large collection of 11 million cell graphs derived from breast tissues and evaluate its transferability across in- and out-of-domain benchmarks. Our results show that GrapHist achieves competitive performance compared to its vision-based counterparts in slide-, region-, and cell-level tasks, while requiring four times fewer parameters. It also drastically outperforms fully-supervised graph models on cancer subtyping tasks. Finally, we also release five graph-based digital pathology datasets used in our study at https://huggingface.co/ogutsevda/datasets , establishing the first large-scale graph benchmark in this field. Our code is available at https://github.com/ogutsevda/graphist .