A Deep Learning Approach for Segmenting the Subplate and Proliferative Zones in Fetal Brain MRI

Helena S. Sousa; Abi Fukami-Gartner; Alena U. Uus; Vanessa Kyriakopoulou; Brigita Ziukaite; Isa Anzak; Jonathan O’Muircheartaigh; Joseph V. Hajnal; J. Donald Tournier; Alexander Hammers; Mary A. Rutherford; Maria Deprez

doi:10.1007/978-3-031-45544-5_2

A Deep Learning Approach for Segmenting the Subplate and Proliferative Zones in Fetal Brain MRI

Helena S. Sousa^*, Abi Fukami-Gartner, Alena U. Uus, Vanessa Kyriakopoulou, Brigita Ziukaite, Isa Anzak, Jonathan O’Muircheartaigh, Joseph V. Hajnal, J. Donald Tournier, Alexander Hammers, Mary A. Rutherford, Maria Deprez

^*Corresponding author for this work

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

Abstract

In vivo fetal brain MRI is employed in clinical practice and in research studies to appreciate in utero brain development. There is increasing interest in transient regions of the fetal brain, such as the subplate (SP), ventricular zone (VZ) and ganglionic eminence (GE) (also referred to as germinal matrix), and their role in normal and abnormal antenatal brain development. On T1w and T2w fetal MRI, these transient regions are defined by highly heterogeneous and stratified signal intensities with rapidly changing patterns. In this work, we define the SP, VZ and GE in a 0.5mm isotropic resolution atlas from the developing Human Connectome Project (dHCP) [1, 17, 18], and train an Attention U-Net [12] to automatically segment them based on semi-automatically generated labels. Our solution spans from 21 through to 36 weeks gestational age (GA), offering insight into a crucial period of antenatal brain development. The proposed automated segmentation achieved mean Dice scores of 0.88, 0.70 and 0.82 for SP, VZ and GE respectively. A volumetric comparison of transient regions in a small cohort of fetuses with isolated ventriculomegaly (VM, n = 8) vs. controls (n = 265) showed significantly enlarged absolute volumes in the GE (P = 0.005) and VZ (P < 0.001) of the left hemisphere.

Original language	English
Title of host publication	Perinatal, Preterm and Paediatric Image Analysis - 8th International Workshop, PIPPI 2023, Held in Conjunction with MICCAI 2023, Proceedings
Editors	Daphna Link-Sourani, Esra Abaci Turk, Christopher Macgowan, Jana Hutter, Jana Hutter, Andrew Melbourne, Roxane Licandro, Roxane Licandro
Publisher	Springer Science and Business Media Deutschland GmbH
Chapter	8
Pages	17-27
Number of pages	11
ISBN (Print)	9783031455438
DOIs	https://doi.org/10.1007/978-3-031-45544-5_2
Publication status	Published - 12 Oct 2023
Event	8th International Workshop on Perinatal, Preterm and Paediatric Image Analysis, PIPPI 2023 - Vancouver, Canada Duration: 12 Oct 2023 → 12 Oct 2023

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	14246 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	8th International Workshop on Perinatal, Preterm and Paediatric Image Analysis, PIPPI 2023
Country/Territory	Canada
City	Vancouver
Period	12/10/2023 → 12/10/2023

Keywords

Automated Segmentation
Brain maturation
Fetal brain MRI
Ganglionic Eminence
Germinal Matrix
Subplate
Ventricular Zone

Access to Document

10.1007/978-3-031-45544-5_2

Cite this

Sousa, H. S., Fukami-Gartner, A., Uus, A. U., Kyriakopoulou, V., Ziukaite, B., Anzak, I., O’Muircheartaigh, J., Hajnal, J. V., Tournier, J. D., Hammers, A., Rutherford, M. A., & Deprez, M. (2023). A Deep Learning Approach for Segmenting the Subplate and Proliferative Zones in Fetal Brain MRI. In D. Link-Sourani, E. Abaci Turk, C. Macgowan, J. Hutter, J. Hutter, A. Melbourne, R. Licandro, & R. Licandro (Eds.), Perinatal, Preterm and Paediatric Image Analysis - 8th International Workshop, PIPPI 2023, Held in Conjunction with MICCAI 2023, Proceedings (pp. 17-27). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14246 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-45544-5_2

Sousa, Helena S. ; Fukami-Gartner, Abi ; Uus, Alena U. et al. / A Deep Learning Approach for Segmenting the Subplate and Proliferative Zones in Fetal Brain MRI. Perinatal, Preterm and Paediatric Image Analysis - 8th International Workshop, PIPPI 2023, Held in Conjunction with MICCAI 2023, Proceedings. editor / Daphna Link-Sourani ; Esra Abaci Turk ; Christopher Macgowan ; Jana Hutter ; Jana Hutter ; Andrew Melbourne ; Roxane Licandro ; Roxane Licandro. Springer Science and Business Media Deutschland GmbH, 2023. pp. 17-27 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inbook{2f4fd150c9e3466285fd84999cbec6d5,

title = "A Deep Learning Approach for Segmenting the Subplate and Proliferative Zones in Fetal Brain MRI",

abstract = "In vivo fetal brain MRI is employed in clinical practice and in research studies to appreciate in utero brain development. There is increasing interest in transient regions of the fetal brain, such as the subplate (SP), ventricular zone (VZ) and ganglionic eminence (GE) (also referred to as germinal matrix), and their role in normal and abnormal antenatal brain development. On T1w and T2w fetal MRI, these transient regions are defined by highly heterogeneous and stratified signal intensities with rapidly changing patterns. In this work, we define the SP, VZ and GE in a 0.5mm isotropic resolution atlas from the developing Human Connectome Project (dHCP) [1, 17, 18], and train an Attention U-Net [12] to automatically segment them based on semi-automatically generated labels. Our solution spans from 21 through to 36 weeks gestational age (GA), offering insight into a crucial period of antenatal brain development. The proposed automated segmentation achieved mean Dice scores of 0.88, 0.70 and 0.82 for SP, VZ and GE respectively. A volumetric comparison of transient regions in a small cohort of fetuses with isolated ventriculomegaly (VM, n = 8) vs. controls (n = 265) showed significantly enlarged absolute volumes in the GE (P = 0.005) and VZ (P < 0.001) of the left hemisphere.",

keywords = "Automated Segmentation, Brain maturation, Fetal brain MRI, Ganglionic Eminence, Germinal Matrix, Subplate, Ventricular Zone",

author = "Sousa, {Helena S.} and Abi Fukami-Gartner and Uus, {Alena U.} and Vanessa Kyriakopoulou and Brigita Ziukaite and Isa Anzak and Jonathan O{\textquoteright}Muircheartaigh and Hajnal, {Joseph V.} and Tournier, {J. Donald} and Alexander Hammers and Rutherford, {Mary A.} and Maria Deprez",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.; 8th International Workshop on Perinatal, Preterm and Paediatric Image Analysis, PIPPI 2023 ; Conference date: 12-10-2023 Through 12-10-2023",

year = "2023",

month = oct,

day = "12",

doi = "10.1007/978-3-031-45544-5_2",

language = "English",

isbn = "9783031455438",

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 = "17--27",

editor = "Daphna Link-Sourani and {Abaci Turk}, Esra and Christopher Macgowan and Jana Hutter and Jana Hutter and Andrew Melbourne and Roxane Licandro and Roxane Licandro",

booktitle = "Perinatal, Preterm and Paediatric Image Analysis - 8th International Workshop, PIPPI 2023, Held in Conjunction with MICCAI 2023, Proceedings",

address = "Germany",

}

Sousa, HS, Fukami-Gartner, A, Uus, AU , Kyriakopoulou, V, Ziukaite, B, Anzak, I, O’Muircheartaigh, J , Hajnal, JV , Tournier, JD , Hammers, A , Rutherford, MA & Deprez, M 2023, A Deep Learning Approach for Segmenting the Subplate and Proliferative Zones in Fetal Brain MRI. in D Link-Sourani, E Abaci Turk, C Macgowan, J Hutter, J Hutter, A Melbourne, R Licandro & R Licandro (eds), Perinatal, Preterm and Paediatric Image Analysis - 8th International Workshop, PIPPI 2023, Held in Conjunction with MICCAI 2023, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 14246 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 17-27, 8th International Workshop on Perinatal, Preterm and Paediatric Image Analysis, PIPPI 2023, Vancouver, Canada, 12/10/2023. https://doi.org/10.1007/978-3-031-45544-5_2

A Deep Learning Approach for Segmenting the Subplate and Proliferative Zones in Fetal Brain MRI. / Sousa, Helena S.; Fukami-Gartner, Abi; Uus, Alena U. et al.
Perinatal, Preterm and Paediatric Image Analysis - 8th International Workshop, PIPPI 2023, Held in Conjunction with MICCAI 2023, Proceedings. ed. / Daphna Link-Sourani; Esra Abaci Turk; Christopher Macgowan; Jana Hutter; Jana Hutter; Andrew Melbourne; Roxane Licandro; Roxane Licandro. Springer Science and Business Media Deutschland GmbH, 2023. p. 17-27 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 14246 LNCS).

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

TY - CHAP

T1 - A Deep Learning Approach for Segmenting the Subplate and Proliferative Zones in Fetal Brain MRI

AU - Sousa, Helena S.

AU - Fukami-Gartner, Abi

AU - Uus, Alena U.

AU - Kyriakopoulou, Vanessa

AU - Ziukaite, Brigita

AU - Anzak, Isa

AU - O’Muircheartaigh, Jonathan

AU - Hajnal, Joseph V.

AU - Tournier, J. Donald

AU - Hammers, Alexander

AU - Rutherford, Mary A.

AU - Deprez, Maria

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.

PY - 2023/10/12

Y1 - 2023/10/12

N2 - In vivo fetal brain MRI is employed in clinical practice and in research studies to appreciate in utero brain development. There is increasing interest in transient regions of the fetal brain, such as the subplate (SP), ventricular zone (VZ) and ganglionic eminence (GE) (also referred to as germinal matrix), and their role in normal and abnormal antenatal brain development. On T1w and T2w fetal MRI, these transient regions are defined by highly heterogeneous and stratified signal intensities with rapidly changing patterns. In this work, we define the SP, VZ and GE in a 0.5mm isotropic resolution atlas from the developing Human Connectome Project (dHCP) [1, 17, 18], and train an Attention U-Net [12] to automatically segment them based on semi-automatically generated labels. Our solution spans from 21 through to 36 weeks gestational age (GA), offering insight into a crucial period of antenatal brain development. The proposed automated segmentation achieved mean Dice scores of 0.88, 0.70 and 0.82 for SP, VZ and GE respectively. A volumetric comparison of transient regions in a small cohort of fetuses with isolated ventriculomegaly (VM, n = 8) vs. controls (n = 265) showed significantly enlarged absolute volumes in the GE (P = 0.005) and VZ (P < 0.001) of the left hemisphere.

AB - In vivo fetal brain MRI is employed in clinical practice and in research studies to appreciate in utero brain development. There is increasing interest in transient regions of the fetal brain, such as the subplate (SP), ventricular zone (VZ) and ganglionic eminence (GE) (also referred to as germinal matrix), and their role in normal and abnormal antenatal brain development. On T1w and T2w fetal MRI, these transient regions are defined by highly heterogeneous and stratified signal intensities with rapidly changing patterns. In this work, we define the SP, VZ and GE in a 0.5mm isotropic resolution atlas from the developing Human Connectome Project (dHCP) [1, 17, 18], and train an Attention U-Net [12] to automatically segment them based on semi-automatically generated labels. Our solution spans from 21 through to 36 weeks gestational age (GA), offering insight into a crucial period of antenatal brain development. The proposed automated segmentation achieved mean Dice scores of 0.88, 0.70 and 0.82 for SP, VZ and GE respectively. A volumetric comparison of transient regions in a small cohort of fetuses with isolated ventriculomegaly (VM, n = 8) vs. controls (n = 265) showed significantly enlarged absolute volumes in the GE (P = 0.005) and VZ (P < 0.001) of the left hemisphere.

KW - Automated Segmentation

KW - Brain maturation

KW - Fetal brain MRI

KW - Ganglionic Eminence

KW - Germinal Matrix

KW - Subplate

KW - Ventricular Zone

UR - http://www.scopus.com/inward/record.url?scp=85175793900&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-45544-5_2

DO - 10.1007/978-3-031-45544-5_2

M3 - Conference paper

AN - SCOPUS:85175793900

SN - 9783031455438

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

SP - 17

EP - 27

BT - Perinatal, Preterm and Paediatric Image Analysis - 8th International Workshop, PIPPI 2023, Held in Conjunction with MICCAI 2023, Proceedings

A2 - Link-Sourani, Daphna

A2 - Abaci Turk, Esra

A2 - Macgowan, Christopher

A2 - Hutter, Jana

A2 - Melbourne, Andrew

A2 - Licandro, Roxane

PB - Springer Science and Business Media Deutschland GmbH

T2 - 8th International Workshop on Perinatal, Preterm and Paediatric Image Analysis, PIPPI 2023

Y2 - 12 October 2023 through 12 October 2023

ER -

Sousa HS, Fukami-Gartner A, Uus AU , Kyriakopoulou V, Ziukaite B, Anzak I et al. A Deep Learning Approach for Segmenting the Subplate and Proliferative Zones in Fetal Brain MRI. In Link-Sourani D, Abaci Turk E, Macgowan C, Hutter J, Hutter J, Melbourne A, Licandro R, Licandro R, editors, Perinatal, Preterm and Paediatric Image Analysis - 8th International Workshop, PIPPI 2023, Held in Conjunction with MICCAI 2023, Proceedings. Springer Science and Business Media Deutschland GmbH. 2023. p. 17-27. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-45544-5_2