Enhancing Cross-Institute Generalisation of GNNs in Histopathology Through Multiple Embedding Graph Augmentation (MEGA)

Jonathan Campbell; Claudia Vanea; Liis Salumäe; Karen Meir; Drorith Hochner-Celnikier; Hagit Hochner; Triin Laisk; Linda M. Ernst; Cecilia M. Lindgren; Weidi Xie; Christoffer Nellåker

doi:10.1007/978-3-031-66958-3_20

Enhancing Cross-Institute Generalisation of GNNs in Histopathology Through Multiple Embedding Graph Augmentation (MEGA)

Jonathan Campbell^*, Claudia Vanea^*, Liis Salumäe, Karen Meir, Drorith Hochner-Celnikier, Hagit Hochner, Triin Laisk, Linda M. Ernst, Cecilia M. Lindgren, Weidi Xie, Christoffer Nellåker^*

^*Corresponding author for this work

Department of Social Medicine

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Many recent methods for the analysis of histology whole slide images (WSIs) have used graph neural networks (GNNs) to aggregate visual information over a large image resolution. However, domain shift is a significant challenge in computational histopathology, due to differences in WSI appearance between institutes, and the effect of these differences on training GNNs has not been explored. In this work, we present the Multiple Embedding Graph Augmentation (MEGA) strategy to improve the cross-institute generalisation of GNNs in histology. We show that by introducing image augmentation and normalisation to the node features used to train a GNN, we can train a model that is robust to domain shift without additional labels or further training of the feature extractor. We compare MEGA to noise-based regularisation and demonstrate its effectiveness in a node classification tissue prediction task in placenta histology.

Original language	English
Title of host publication	Medical Image Understanding and Analysis - 28th Annual Conference, MIUA 2024, Proceedings
Editors	Moi Hoon Yap, Connah Kendrick, Ardhendu Behera, Timothy Cootes, Reyer Zwiggelaar
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	270-284
Number of pages	15
ISBN (Print)	9783031669576
DOIs	https://doi.org/10.1007/978-3-031-66958-3_20
State	Published - 2024
Event	28th Annual Conference on Medical Image Understanding and Analysis, MIUA 2024 - Manchester, United Kingdom Duration: 24 Jul 2024 → 26 Jul 2024

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	14860 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	28th Annual Conference on Medical Image Understanding and Analysis, MIUA 2024
Country/Territory	United Kingdom
City	Manchester
Period	24/07/24 → 26/07/24

Bibliographical note

Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.

Keywords

Deep Learning
Domain Generalisation
Graph Neural Networks
Histology

Access to Document

10.1007/978-3-031-66958-3_20

Cite this

Campbell, J., Vanea, C., Salumäe, L., Meir, K., Hochner-Celnikier, D., Hochner, H., Laisk, T., Ernst, L. M., Lindgren, C. M., Xie, W., & Nellåker, C. (2024). Enhancing Cross-Institute Generalisation of GNNs in Histopathology Through Multiple Embedding Graph Augmentation (MEGA). In M. H. Yap, C. Kendrick, A. Behera, T. Cootes, & R. Zwiggelaar (Eds.), Medical Image Understanding and Analysis - 28th Annual Conference, MIUA 2024, Proceedings (pp. 270-284). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14860 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-66958-3_20

Campbell, Jonathan ; Vanea, Claudia ; Salumäe, Liis et al. / Enhancing Cross-Institute Generalisation of GNNs in Histopathology Through Multiple Embedding Graph Augmentation (MEGA). Medical Image Understanding and Analysis - 28th Annual Conference, MIUA 2024, Proceedings. editor / Moi Hoon Yap ; Connah Kendrick ; Ardhendu Behera ; Timothy Cootes ; Reyer Zwiggelaar. Springer Science and Business Media Deutschland GmbH, 2024. pp. 270-284 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{e4dc680c3ab746048466af4d08d7db56,

title = "Enhancing Cross-Institute Generalisation of GNNs in Histopathology Through Multiple Embedding Graph Augmentation (MEGA)",

abstract = "Many recent methods for the analysis of histology whole slide images (WSIs) have used graph neural networks (GNNs) to aggregate visual information over a large image resolution. However, domain shift is a significant challenge in computational histopathology, due to differences in WSI appearance between institutes, and the effect of these differences on training GNNs has not been explored. In this work, we present the Multiple Embedding Graph Augmentation (MEGA) strategy to improve the cross-institute generalisation of GNNs in histology. We show that by introducing image augmentation and normalisation to the node features used to train a GNN, we can train a model that is robust to domain shift without additional labels or further training of the feature extractor. We compare MEGA to noise-based regularisation and demonstrate its effectiveness in a node classification tissue prediction task in placenta histology.",

keywords = "Deep Learning, Domain Generalisation, Graph Neural Networks, Histology",

author = "Jonathan Campbell and Claudia Vanea and Liis Salum{\"a}e and Karen Meir and Drorith Hochner-Celnikier and Hagit Hochner and Triin Laisk and Ernst, {Linda M.} and Lindgren, {Cecilia M.} and Weidi Xie and Christoffer Nell{\aa}ker",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.; 28th Annual Conference on Medical Image Understanding and Analysis, MIUA 2024 ; Conference date: 24-07-2024 Through 26-07-2024",

year = "2024",

doi = "10.1007/978-3-031-66958-3_20",

language = "אנגלית",

isbn = "9783031669576",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "270--284",

editor = "Yap, {Moi Hoon} and Connah Kendrick and Ardhendu Behera and Timothy Cootes and Reyer Zwiggelaar",

booktitle = "Medical Image Understanding and Analysis - 28th Annual Conference, MIUA 2024, Proceedings",

address = "גרמניה",

}

Campbell, J, Vanea, C, Salumäe, L, Meir, K, Hochner-Celnikier, D, Hochner, H, Laisk, T, Ernst, LM, Lindgren, CM, Xie, W & Nellåker, C 2024, Enhancing Cross-Institute Generalisation of GNNs in Histopathology Through Multiple Embedding Graph Augmentation (MEGA). in MH Yap, C Kendrick, A Behera, T Cootes & R Zwiggelaar (eds), Medical Image Understanding and Analysis - 28th Annual Conference, MIUA 2024, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 14860 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 270-284, 28th Annual Conference on Medical Image Understanding and Analysis, MIUA 2024, Manchester, United Kingdom, 24/07/24. https://doi.org/10.1007/978-3-031-66958-3_20

Enhancing Cross-Institute Generalisation of GNNs in Histopathology Through Multiple Embedding Graph Augmentation (MEGA). / Campbell, Jonathan; Vanea, Claudia; Salumäe, Liis et al.
Medical Image Understanding and Analysis - 28th Annual Conference, MIUA 2024, Proceedings. ed. / Moi Hoon Yap; Connah Kendrick; Ardhendu Behera; Timothy Cootes; Reyer Zwiggelaar. Springer Science and Business Media Deutschland GmbH, 2024. p. 270-284 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14860 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Enhancing Cross-Institute Generalisation of GNNs in Histopathology Through Multiple Embedding Graph Augmentation (MEGA)

AU - Campbell, Jonathan

AU - Vanea, Claudia

AU - Salumäe, Liis

AU - Meir, Karen

AU - Hochner-Celnikier, Drorith

AU - Hochner, Hagit

AU - Laisk, Triin

AU - Ernst, Linda M.

AU - Lindgren, Cecilia M.

AU - Xie, Weidi

AU - Nellåker, Christoffer

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.

PY - 2024

Y1 - 2024

N2 - Many recent methods for the analysis of histology whole slide images (WSIs) have used graph neural networks (GNNs) to aggregate visual information over a large image resolution. However, domain shift is a significant challenge in computational histopathology, due to differences in WSI appearance between institutes, and the effect of these differences on training GNNs has not been explored. In this work, we present the Multiple Embedding Graph Augmentation (MEGA) strategy to improve the cross-institute generalisation of GNNs in histology. We show that by introducing image augmentation and normalisation to the node features used to train a GNN, we can train a model that is robust to domain shift without additional labels or further training of the feature extractor. We compare MEGA to noise-based regularisation and demonstrate its effectiveness in a node classification tissue prediction task in placenta histology.

AB - Many recent methods for the analysis of histology whole slide images (WSIs) have used graph neural networks (GNNs) to aggregate visual information over a large image resolution. However, domain shift is a significant challenge in computational histopathology, due to differences in WSI appearance between institutes, and the effect of these differences on training GNNs has not been explored. In this work, we present the Multiple Embedding Graph Augmentation (MEGA) strategy to improve the cross-institute generalisation of GNNs in histology. We show that by introducing image augmentation and normalisation to the node features used to train a GNN, we can train a model that is robust to domain shift without additional labels or further training of the feature extractor. We compare MEGA to noise-based regularisation and demonstrate its effectiveness in a node classification tissue prediction task in placenta histology.

KW - Deep Learning

KW - Domain Generalisation

KW - Graph Neural Networks

KW - Histology

UR - http://www.scopus.com/inward/record.url?scp=85200675099&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-66958-3_20

DO - 10.1007/978-3-031-66958-3_20

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

AN - SCOPUS:85200675099

SN - 9783031669576

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 270

EP - 284

BT - Medical Image Understanding and Analysis - 28th Annual Conference, MIUA 2024, Proceedings

A2 - Yap, Moi Hoon

A2 - Kendrick, Connah

A2 - Behera, Ardhendu

A2 - Cootes, Timothy

A2 - Zwiggelaar, Reyer

PB - Springer Science and Business Media Deutschland GmbH

T2 - 28th Annual Conference on Medical Image Understanding and Analysis, MIUA 2024

Y2 - 24 July 2024 through 26 July 2024

ER -

Campbell J, Vanea C, Salumäe L, Meir K, Hochner-Celnikier D, Hochner H et al. Enhancing Cross-Institute Generalisation of GNNs in Histopathology Through Multiple Embedding Graph Augmentation (MEGA). In Yap MH, Kendrick C, Behera A, Cootes T, Zwiggelaar R, editors, Medical Image Understanding and Analysis - 28th Annual Conference, MIUA 2024, Proceedings. Springer Science and Business Media Deutschland GmbH. 2024. p. 270-284. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-66958-3_20