Harmonization of infant cortical thickness using surface-to-surface cycle-consistent adversarial networks

the UNC/UMN Baby Connectome Project Consortium

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Increasing multi-site infant neuroimaging datasets are facilitating the research on understanding early brain development with larger sample size and bigger statistical power. However, a joint analysis of cortical properties (e.g., cortical thickness) is unavoidably facing the problem of non-biological variance introduced by differences in MRI scanners. To address this issue, in this paper, we propose cycle-consistent adversarial networks based on spherical cortical surface to harmonize cortical thickness maps between different scanners. We combine the spherical U-Net and CycleGAN to construct a surface-to-surface CycleGAN (S2SGAN). Specifically, we model the harmonization from scanner X to scanner Y as a surface-to-surface translation task. The first goal of harmonization is to learn a mapping X→Y such that the distribution of surface thickness maps from GX(X) is indistinguishable from Y. Since this mapping is highly under-constrained, with the second goal of harmonization to preserve individual differences, we utilize the inverse mapping Gy:Y→X and the cycle consistency loss to enforce GY (GX(X)) ≈ X (and vice versa). Furthermore, we incorporate the correlation coefficient loss to guarantee the structure consistency between the original and the generated surface thickness maps. Quantitative evaluation on both synthesized and real infant cortical data demonstrates the superior ability of our method in removing unwanted scanner effects and preserving individual differences simultaneously, compared to the state-of-the-art methods.

Original languageEnglish
Title of host publicationMedical Image Computing and Computer Assisted Intervention – MICCAI 2019 - 22nd International Conference, Proceedings
EditorsDinggang Shen, Pew-Thian Yap, Tianming Liu, Terry M. Peters, Ali Khan, Lawrence H. Staib, Caroline Essert, Sean Zhou
PublisherSpringer
Pages475-483
Number of pages9
ISBN (Print)9783030322502
DOIs
Publication statusPublished - 2019 Jan 1
Externally publishedYes
Event22nd International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2019 - Shenzhen, China
Duration: 2019 Oct 132019 Oct 17

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11767 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference22nd International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2019
CountryChina
CityShenzhen
Period19/10/1319/10/17

Fingerprint

Scanner
Cycle
Individual Differences
Neuroimaging
Statistical Power
Quantitative Evaluation
Correlation coefficient
Magnetic resonance imaging
Brain
Sample Size
Demonstrate

Keywords

  • CycleGAN
  • Harmonization
  • Spherical U-Net

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

the UNC/UMN Baby Connectome Project Consortium (2019). Harmonization of infant cortical thickness using surface-to-surface cycle-consistent adversarial networks. In D. Shen, P-T. Yap, T. Liu, T. M. Peters, A. Khan, L. H. Staib, C. Essert, ... S. Zhou (Eds.), Medical Image Computing and Computer Assisted Intervention – MICCAI 2019 - 22nd International Conference, Proceedings (pp. 475-483). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11767 LNCS). Springer. https://doi.org/10.1007/978-3-030-32251-9_52

Harmonization of infant cortical thickness using surface-to-surface cycle-consistent adversarial networks. / the UNC/UMN Baby Connectome Project Consortium.

Medical Image Computing and Computer Assisted Intervention – MICCAI 2019 - 22nd International Conference, Proceedings. ed. / Dinggang Shen; Pew-Thian Yap; Tianming Liu; Terry M. Peters; Ali Khan; Lawrence H. Staib; Caroline Essert; Sean Zhou. Springer, 2019. p. 475-483 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11767 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

the UNC/UMN Baby Connectome Project Consortium 2019, Harmonization of infant cortical thickness using surface-to-surface cycle-consistent adversarial networks. in D Shen, P-T Yap, T Liu, TM Peters, A Khan, LH Staib, C Essert & S Zhou (eds), Medical Image Computing and Computer Assisted Intervention – MICCAI 2019 - 22nd International Conference, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11767 LNCS, Springer, pp. 475-483, 22nd International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI 2019, Shenzhen, China, 19/10/13. https://doi.org/10.1007/978-3-030-32251-9_52
the UNC/UMN Baby Connectome Project Consortium. Harmonization of infant cortical thickness using surface-to-surface cycle-consistent adversarial networks. In Shen D, Yap P-T, Liu T, Peters TM, Khan A, Staib LH, Essert C, Zhou S, editors, Medical Image Computing and Computer Assisted Intervention – MICCAI 2019 - 22nd International Conference, Proceedings. Springer. 2019. p. 475-483. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-32251-9_52
the UNC/UMN Baby Connectome Project Consortium. / Harmonization of infant cortical thickness using surface-to-surface cycle-consistent adversarial networks. Medical Image Computing and Computer Assisted Intervention – MICCAI 2019 - 22nd International Conference, Proceedings. editor / Dinggang Shen ; Pew-Thian Yap ; Tianming Liu ; Terry M. Peters ; Ali Khan ; Lawrence H. Staib ; Caroline Essert ; Sean Zhou. Springer, 2019. pp. 475-483 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{b5f0b8366636465fa0b94b683ba1bdd8,
title = "Harmonization of infant cortical thickness using surface-to-surface cycle-consistent adversarial networks",
abstract = "Increasing multi-site infant neuroimaging datasets are facilitating the research on understanding early brain development with larger sample size and bigger statistical power. However, a joint analysis of cortical properties (e.g., cortical thickness) is unavoidably facing the problem of non-biological variance introduced by differences in MRI scanners. To address this issue, in this paper, we propose cycle-consistent adversarial networks based on spherical cortical surface to harmonize cortical thickness maps between different scanners. We combine the spherical U-Net and CycleGAN to construct a surface-to-surface CycleGAN (S2SGAN). Specifically, we model the harmonization from scanner X to scanner Y as a surface-to-surface translation task. The first goal of harmonization is to learn a mapping X→Y such that the distribution of surface thickness maps from GX(X) is indistinguishable from Y. Since this mapping is highly under-constrained, with the second goal of harmonization to preserve individual differences, we utilize the inverse mapping Gy:Y→X and the cycle consistency loss to enforce GY (GX(X)) ≈ X (and vice versa). Furthermore, we incorporate the correlation coefficient loss to guarantee the structure consistency between the original and the generated surface thickness maps. Quantitative evaluation on both synthesized and real infant cortical data demonstrates the superior ability of our method in removing unwanted scanner effects and preserving individual differences simultaneously, compared to the state-of-the-art methods.",
keywords = "CycleGAN, Harmonization, Spherical U-Net",
author = "{the UNC/UMN Baby Connectome Project Consortium} and Fenqiang Zhao and Zhengwang Wu and Li Wang and Weili Lin and Shunren Xia and Dinggang Shen and Gang Li",
year = "2019",
month = "1",
day = "1",
doi = "10.1007/978-3-030-32251-9_52",
language = "English",
isbn = "9783030322502",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer",
pages = "475--483",
editor = "Dinggang Shen and Pew-Thian Yap and Tianming Liu and Peters, {Terry M.} and Ali Khan and Staib, {Lawrence H.} and Caroline Essert and Sean Zhou",
booktitle = "Medical Image Computing and Computer Assisted Intervention – MICCAI 2019 - 22nd International Conference, Proceedings",

}

TY - GEN

T1 - Harmonization of infant cortical thickness using surface-to-surface cycle-consistent adversarial networks

AU - the UNC/UMN Baby Connectome Project Consortium

AU - Zhao, Fenqiang

AU - Wu, Zhengwang

AU - Wang, Li

AU - Lin, Weili

AU - Xia, Shunren

AU - Shen, Dinggang

AU - Li, Gang

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Increasing multi-site infant neuroimaging datasets are facilitating the research on understanding early brain development with larger sample size and bigger statistical power. However, a joint analysis of cortical properties (e.g., cortical thickness) is unavoidably facing the problem of non-biological variance introduced by differences in MRI scanners. To address this issue, in this paper, we propose cycle-consistent adversarial networks based on spherical cortical surface to harmonize cortical thickness maps between different scanners. We combine the spherical U-Net and CycleGAN to construct a surface-to-surface CycleGAN (S2SGAN). Specifically, we model the harmonization from scanner X to scanner Y as a surface-to-surface translation task. The first goal of harmonization is to learn a mapping X→Y such that the distribution of surface thickness maps from GX(X) is indistinguishable from Y. Since this mapping is highly under-constrained, with the second goal of harmonization to preserve individual differences, we utilize the inverse mapping Gy:Y→X and the cycle consistency loss to enforce GY (GX(X)) ≈ X (and vice versa). Furthermore, we incorporate the correlation coefficient loss to guarantee the structure consistency between the original and the generated surface thickness maps. Quantitative evaluation on both synthesized and real infant cortical data demonstrates the superior ability of our method in removing unwanted scanner effects and preserving individual differences simultaneously, compared to the state-of-the-art methods.

AB - Increasing multi-site infant neuroimaging datasets are facilitating the research on understanding early brain development with larger sample size and bigger statistical power. However, a joint analysis of cortical properties (e.g., cortical thickness) is unavoidably facing the problem of non-biological variance introduced by differences in MRI scanners. To address this issue, in this paper, we propose cycle-consistent adversarial networks based on spherical cortical surface to harmonize cortical thickness maps between different scanners. We combine the spherical U-Net and CycleGAN to construct a surface-to-surface CycleGAN (S2SGAN). Specifically, we model the harmonization from scanner X to scanner Y as a surface-to-surface translation task. The first goal of harmonization is to learn a mapping X→Y such that the distribution of surface thickness maps from GX(X) is indistinguishable from Y. Since this mapping is highly under-constrained, with the second goal of harmonization to preserve individual differences, we utilize the inverse mapping Gy:Y→X and the cycle consistency loss to enforce GY (GX(X)) ≈ X (and vice versa). Furthermore, we incorporate the correlation coefficient loss to guarantee the structure consistency between the original and the generated surface thickness maps. Quantitative evaluation on both synthesized and real infant cortical data demonstrates the superior ability of our method in removing unwanted scanner effects and preserving individual differences simultaneously, compared to the state-of-the-art methods.

KW - CycleGAN

KW - Harmonization

KW - Spherical U-Net

UR - http://www.scopus.com/inward/record.url?scp=85075685779&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85075685779&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-32251-9_52

DO - 10.1007/978-3-030-32251-9_52

M3 - Conference contribution

AN - SCOPUS:85075685779

SN - 9783030322502

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

SP - 475

EP - 483

BT - Medical Image Computing and Computer Assisted Intervention – MICCAI 2019 - 22nd International Conference, Proceedings

A2 - Shen, Dinggang

A2 - Yap, Pew-Thian

A2 - Liu, Tianming

A2 - Peters, Terry M.

A2 - Khan, Ali

A2 - Staib, Lawrence H.

A2 - Essert, Caroline

A2 - Zhou, Sean

PB - Springer

ER -