Domain Game: Disentangle Anatomical Feature for Single Domain Generalized Segmentation

Hao Chen; Hongrun Zhang; U. Wang Chan; Rui Yin; Xiaofei Wang; Chao Li

doi:10.1007/978-3-031-73360-4_5

Domain Game: Disentangle Anatomical Feature for Single Domain Generalized Segmentation

Hao Chen, Hongrun Zhang, U. Wang Chan, Rui Yin, Xiaofei Wang, Chao Li (Lead / Corresponding author)

Biomedical Engineering

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

Abstract

Single domain generalization aims to address the challenge of out-of-distribution generalization problem with only one source domain available. Feature disentanglement is a classic solution to this purpose, where the extracted task-related feature is presumed to be resilient to domain shift. However, the absence of references from other domains in a single-domain scenario poses significant uncertainty in feature disentanglement (ill-posedness). In this paper, we propose a new framework, named Domain Game, to perform feature disentangling for medical image segmentation, based on the observation that anatomical features are more sensitive to geometric transformations, whilst domain-specific features probably will remain invariant to such operations. Results from cross-site test domain evaluation showcase approximately an ∼11.8% performance boost in prostate segmentation and around ∼10.5% in brain tumor segmentation. The codes will be available at https://github.com/chqwer2/Domain-Game.

Original language	English
Title of host publication	Computational Mathematics Modeling in Cancer Analysis
Subtitle of host publication	3rd International Workshop, CMMCA 2024, Proceedings
Editors	Jia Wu, Wenjian Qin, Chao Li, Boklye Kim
Publisher	Springer
Pages	41-51
Number of pages	11
ISBN (Electronic)	9783031733604
ISBN (Print)	9783031733598
DOIs	https://doi.org/10.1007/978-3-031-73360-4_5
Publication status	Published - 5 Oct 2024
Event	3rd Workshop on Computational Mathematics Modeling in Cancer Analysis - Marrakesh, Morocco Duration: 6 Oct 2024 → 6 Oct 2024 https://cmmcaworkshop.github.io/2024/ (Link to Conference)

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	15181 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	3rd Workshop on Computational Mathematics Modeling in Cancer Analysis
Abbreviated title	CMMCA 2024
Country/Territory	Morocco
City	Marrakesh
Period	6/10/24 → 6/10/24
Other	CMMCA 2024 was held in conjunction with the 27th International Conference on Medical Image Computing and Computer Assisted Intervention, MICCAI 2024
Internet address	https://cmmcaworkshop.github.io/2024/ (Link to Conference)

Keywords

Feature disentanglement
Generalization
Segmentation

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Access to Document

10.1007/978-3-031-73360-4_5

Cite this

Chen, H., Zhang, H., Chan, U. W., Yin, R., Wang, X., & Li, C. (2024). Domain Game: Disentangle Anatomical Feature for Single Domain Generalized Segmentation. In J. Wu, W. Qin, C. Li, & B. Kim (Eds.), Computational Mathematics Modeling in Cancer Analysis: 3rd International Workshop, CMMCA 2024, Proceedings (pp. 41-51). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 15181 LNCS). Springer . https://doi.org/10.1007/978-3-031-73360-4_5

Chen, Hao ; Zhang, Hongrun ; Chan, U. Wang et al. / Domain Game : Disentangle Anatomical Feature for Single Domain Generalized Segmentation. Computational Mathematics Modeling in Cancer Analysis: 3rd International Workshop, CMMCA 2024, Proceedings. editor / Jia Wu ; Wenjian Qin ; Chao Li ; Boklye Kim. Springer , 2024. pp. 41-51 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{a359e71ecce9441c9440f49e8f7f523e,

title = "Domain Game: Disentangle Anatomical Feature for Single Domain Generalized Segmentation",

abstract = "Single domain generalization aims to address the challenge of out-of-distribution generalization problem with only one source domain available. Feature disentanglement is a classic solution to this purpose, where the extracted task-related feature is presumed to be resilient to domain shift. However, the absence of references from other domains in a single-domain scenario poses significant uncertainty in feature disentanglement (ill-posedness). In this paper, we propose a new framework, named Domain Game, to perform feature disentangling for medical image segmentation, based on the observation that anatomical features are more sensitive to geometric transformations, whilst domain-specific features probably will remain invariant to such operations. Results from cross-site test domain evaluation showcase approximately an ∼11.8% performance boost in prostate segmentation and around ∼10.5% in brain tumor segmentation. The codes will be available at https://github.com/chqwer2/Domain-Game.",

keywords = "Feature disentanglement, Generalization, Segmentation",

author = "Hao Chen and Hongrun Zhang and Chan, {U. Wang} and Rui Yin and Xiaofei Wang and Chao Li",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.; 3rd Workshop on Computational Mathematics Modeling in Cancer Analysis, CMMCA 2024 ; Conference date: 06-10-2024 Through 06-10-2024",

year = "2024",

month = oct,

day = "5",

doi = "10.1007/978-3-031-73360-4_5",

language = "English",

isbn = "9783031733598",

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

publisher = "Springer ",

pages = "41--51",

editor = "Jia Wu and Wenjian Qin and Chao Li and Boklye Kim",

booktitle = "Computational Mathematics Modeling in Cancer Analysis",

url = "https://cmmcaworkshop.github.io/2024/",

}

Chen, H, Zhang, H, Chan, UW, Yin, R, Wang, X & Li, C 2024, Domain Game: Disentangle Anatomical Feature for Single Domain Generalized Segmentation. in J Wu, W Qin, C Li & B Kim (eds), Computational Mathematics Modeling in Cancer Analysis: 3rd International Workshop, CMMCA 2024, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 15181 LNCS, Springer , pp. 41-51, 3rd Workshop on Computational Mathematics Modeling in Cancer Analysis, Marrakesh, Morocco, 6/10/24. https://doi.org/10.1007/978-3-031-73360-4_5

Domain Game: Disentangle Anatomical Feature for Single Domain Generalized Segmentation. / Chen, Hao; Zhang, Hongrun; Chan, U. Wang et al.
Computational Mathematics Modeling in Cancer Analysis: 3rd International Workshop, CMMCA 2024, Proceedings. ed. / Jia Wu; Wenjian Qin; Chao Li; Boklye Kim. Springer , 2024. p. 41-51 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 15181 LNCS).

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

TY - GEN

T1 - Domain Game

T2 - 3rd Workshop on Computational Mathematics Modeling in Cancer Analysis

AU - Chen, Hao

AU - Zhang, Hongrun

AU - Chan, U. Wang

AU - Yin, Rui

AU - Wang, Xiaofei

AU - Li, Chao

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

PY - 2024/10/5

Y1 - 2024/10/5

N2 - Single domain generalization aims to address the challenge of out-of-distribution generalization problem with only one source domain available. Feature disentanglement is a classic solution to this purpose, where the extracted task-related feature is presumed to be resilient to domain shift. However, the absence of references from other domains in a single-domain scenario poses significant uncertainty in feature disentanglement (ill-posedness). In this paper, we propose a new framework, named Domain Game, to perform feature disentangling for medical image segmentation, based on the observation that anatomical features are more sensitive to geometric transformations, whilst domain-specific features probably will remain invariant to such operations. Results from cross-site test domain evaluation showcase approximately an ∼11.8% performance boost in prostate segmentation and around ∼10.5% in brain tumor segmentation. The codes will be available at https://github.com/chqwer2/Domain-Game.

AB - Single domain generalization aims to address the challenge of out-of-distribution generalization problem with only one source domain available. Feature disentanglement is a classic solution to this purpose, where the extracted task-related feature is presumed to be resilient to domain shift. However, the absence of references from other domains in a single-domain scenario poses significant uncertainty in feature disentanglement (ill-posedness). In this paper, we propose a new framework, named Domain Game, to perform feature disentangling for medical image segmentation, based on the observation that anatomical features are more sensitive to geometric transformations, whilst domain-specific features probably will remain invariant to such operations. Results from cross-site test domain evaluation showcase approximately an ∼11.8% performance boost in prostate segmentation and around ∼10.5% in brain tumor segmentation. The codes will be available at https://github.com/chqwer2/Domain-Game.

KW - Feature disentanglement

KW - Generalization

KW - Segmentation

UR - http://www.scopus.com/inward/record.url?scp=85206454609&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-73360-4_5

DO - 10.1007/978-3-031-73360-4_5

M3 - Conference contribution

AN - SCOPUS:85206454609

SN - 9783031733598

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

SP - 41

EP - 51

BT - Computational Mathematics Modeling in Cancer Analysis

A2 - Wu, Jia

A2 - Qin, Wenjian

A2 - Li, Chao

A2 - Kim, Boklye

PB - Springer

Y2 - 6 October 2024 through 6 October 2024

ER -

Chen H, Zhang H, Chan UW, Yin R, Wang X, Li C. Domain Game: Disentangle Anatomical Feature for Single Domain Generalized Segmentation. In Wu J, Qin W, Li C, Kim B, editors, Computational Mathematics Modeling in Cancer Analysis: 3rd International Workshop, CMMCA 2024, Proceedings. Springer . 2024. p. 41-51. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-73360-4_5