Domain Adaptation for Automatic Aorta Segmentation of 4D Flow Magnetic Resonance Imaging Data from Multiple Vendor Scanners

Jordina Aviles; Gonzalo D.Maso Talou; Oscar Camara; Marcos Mejía Córdova; Xabier Morales Ferez; Daniel Romero; Edward Ferdian; Kathleen Gilbert; Ayah Elsayed; Alistair A. Young; Lydia Dux-Santoy; Aroa Ruiz-Munoz; Gisela Teixido-Tura; Jose Rodriguez-Palomares; Andrea Guala

doi:10.1007/978-3-030-78710-3_12

Domain Adaptation for Automatic Aorta Segmentation of 4D Flow Magnetic Resonance Imaging Data from Multiple Vendor Scanners

Jordina Aviles, Gonzalo D.Maso Talou, Oscar Camara^*, Marcos Mejía Córdova, Xabier Morales Ferez, Daniel Romero, Edward Ferdian, Kathleen Gilbert, Ayah Elsayed, Alistair A. Young, Lydia Dux-Santoy, Aroa Ruiz-Munoz, Gisela Teixido-Tura, Jose Rodriguez-Palomares, Andrea Guala

^*Corresponding author for this work

Biomedical Engineering Department

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

8 Citations (Scopus)

Abstract

The lack of standardized pipelines for image processing has prevented the application of deep learning (DL) techniques for the segmentation of the aorta in phase-contrast enhanced magnetic resonance angiography (PC-MRA). Furthermore, large, well-curated and annotated datasets, which are needed to create DL-based models able to generalize, are rare. We present the adaptation of the popular nnU-net DL framework to automatically segment the aorta in 4D flow MRI-derived angiograms. The resulting segmentations in a large database (> 300 cases) with normal cases and examples of different pathologies of the aorta provided from a single centre were excellent after post-processing (Dice score of 0.944). Subsequently, we explored the generalisation of the trained network in a small dataset of images (around 20 cases) acquired in a different hospital with another scanner. Without domain adaptation, only with a model trained with the large dataset, the obtained results were substantially worst than with adding a few cases of the small dataset (Dice scores of 0.61 vs 0.86, respectively). The obtained results created good quality segmentations of the aorta in 4D flow MRI, which can later be post-processed to assess blood flow patterns, similarly than with manual annotations. However, advanced domain adaptation schemes are very important in 4D flow MRI due to the large differences in image characteristics between different vendor scanners available in multiple centers.

Original language	English
Title of host publication	Functional Imaging and Modeling of the Heart - 11th International Conference, FIMH 2021, Proceedings
Editors	Daniel B. Ennis, Luigi E. Perotti, Vicky Y. Wang
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	112-121
Number of pages	10
ISBN (Print)	9783030787097
DOIs	https://doi.org/10.1007/978-3-030-78710-3_12
Publication status	Published - 2021
Event	11th International Conference on Functional Imaging and Modeling of the Heart, FIMH 2021 - Virtual, Online Duration: 21 Jun 2021 → 25 Jun 2021

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12738 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	11th International Conference on Functional Imaging and Modeling of the Heart, FIMH 2021
City	Virtual, Online
Period	21/06/2021 → 25/06/2021

Keywords

4D flow magnetic resonance imaging
Aortic segmentation
Deep learning
nnU-net

Access to Document

10.1007/978-3-030-78710-3_12

Cite this

Aviles, J., Talou, G. D. M., Camara, O., Córdova, M. M., Ferez, X. M., Romero, D., Ferdian, E., Gilbert, K., Elsayed, A., Young, A. A., Dux-Santoy, L., Ruiz-Munoz, A., Teixido-Tura, G., Rodriguez-Palomares, J., & Guala, A. (2021). Domain Adaptation for Automatic Aorta Segmentation of 4D Flow Magnetic Resonance Imaging Data from Multiple Vendor Scanners. In D. B. Ennis, L. E. Perotti, & V. Y. Wang (Eds.), Functional Imaging and Modeling of the Heart - 11th International Conference, FIMH 2021, Proceedings (pp. 112-121). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12738 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-78710-3_12

Aviles, Jordina ; Talou, Gonzalo D.Maso ; Camara, Oscar et al. / Domain Adaptation for Automatic Aorta Segmentation of 4D Flow Magnetic Resonance Imaging Data from Multiple Vendor Scanners. Functional Imaging and Modeling of the Heart - 11th International Conference, FIMH 2021, Proceedings. editor / Daniel B. Ennis ; Luigi E. Perotti ; Vicky Y. Wang. Springer Science and Business Media Deutschland GmbH, 2021. pp. 112-121 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inbook{b838c8638e2c4c57ad8c25b062179fd4,

title = "Domain Adaptation for Automatic Aorta Segmentation of 4D Flow Magnetic Resonance Imaging Data from Multiple Vendor Scanners",

abstract = "The lack of standardized pipelines for image processing has prevented the application of deep learning (DL) techniques for the segmentation of the aorta in phase-contrast enhanced magnetic resonance angiography (PC-MRA). Furthermore, large, well-curated and annotated datasets, which are needed to create DL-based models able to generalize, are rare. We present the adaptation of the popular nnU-net DL framework to automatically segment the aorta in 4D flow MRI-derived angiograms. The resulting segmentations in a large database (> 300 cases) with normal cases and examples of different pathologies of the aorta provided from a single centre were excellent after post-processing (Dice score of 0.944). Subsequently, we explored the generalisation of the trained network in a small dataset of images (around 20 cases) acquired in a different hospital with another scanner. Without domain adaptation, only with a model trained with the large dataset, the obtained results were substantially worst than with adding a few cases of the small dataset (Dice scores of 0.61 vs 0.86, respectively). The obtained results created good quality segmentations of the aorta in 4D flow MRI, which can later be post-processed to assess blood flow patterns, similarly than with manual annotations. However, advanced domain adaptation schemes are very important in 4D flow MRI due to the large differences in image characteristics between different vendor scanners available in multiple centers.",

keywords = "4D flow magnetic resonance imaging, Aortic segmentation, Deep learning, nnU-net",

author = "Jordina Aviles and Talou, {Gonzalo D.Maso} and Oscar Camara and C{\'o}rdova, {Marcos Mej{\'i}a} and Ferez, {Xabier Morales} and Daniel Romero and Edward Ferdian and Kathleen Gilbert and Ayah Elsayed and Young, {Alistair A.} and Lydia Dux-Santoy and Aroa Ruiz-Munoz and Gisela Teixido-Tura and Jose Rodriguez-Palomares and Andrea Guala",

note = "Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG. Copyright: Copyright 2021 Elsevier B.V., All rights reserved.; 11th International Conference on Functional Imaging and Modeling of the Heart, FIMH 2021 ; Conference date: 21-06-2021 Through 25-06-2021",

year = "2021",

doi = "10.1007/978-3-030-78710-3_12",

language = "English",

isbn = "9783030787097",

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 = "112--121",

editor = "Ennis, {Daniel B.} and Perotti, {Luigi E.} and Wang, {Vicky Y.}",

booktitle = "Functional Imaging and Modeling of the Heart - 11th International Conference, FIMH 2021, Proceedings",

address = "Germany",

}

Aviles, J, Talou, GDM, Camara, O, Córdova, MM, Ferez, XM, Romero, D, Ferdian, E, Gilbert, K, Elsayed, A, Young, AA, Dux-Santoy, L, Ruiz-Munoz, A, Teixido-Tura, G, Rodriguez-Palomares, J & Guala, A 2021, Domain Adaptation for Automatic Aorta Segmentation of 4D Flow Magnetic Resonance Imaging Data from Multiple Vendor Scanners. in DB Ennis, LE Perotti & VY Wang (eds), Functional Imaging and Modeling of the Heart - 11th International Conference, FIMH 2021, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12738 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 112-121, 11th International Conference on Functional Imaging and Modeling of the Heart, FIMH 2021, Virtual, Online, 21/06/2021. https://doi.org/10.1007/978-3-030-78710-3_12

Domain Adaptation for Automatic Aorta Segmentation of 4D Flow Magnetic Resonance Imaging Data from Multiple Vendor Scanners. / Aviles, Jordina; Talou, Gonzalo D.Maso; Camara, Oscar et al.
Functional Imaging and Modeling of the Heart - 11th International Conference, FIMH 2021, Proceedings. ed. / Daniel B. Ennis; Luigi E. Perotti; Vicky Y. Wang. Springer Science and Business Media Deutschland GmbH, 2021. p. 112-121 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12738 LNCS).

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

TY - CHAP

T1 - Domain Adaptation for Automatic Aorta Segmentation of 4D Flow Magnetic Resonance Imaging Data from Multiple Vendor Scanners

AU - Aviles, Jordina

AU - Talou, Gonzalo D.Maso

AU - Camara, Oscar

AU - Córdova, Marcos Mejía

AU - Ferez, Xabier Morales

AU - Romero, Daniel

AU - Ferdian, Edward

AU - Gilbert, Kathleen

AU - Elsayed, Ayah

AU - Young, Alistair A.

AU - Dux-Santoy, Lydia

AU - Ruiz-Munoz, Aroa

AU - Teixido-Tura, Gisela

AU - Rodriguez-Palomares, Jose

AU - Guala, Andrea

PY - 2021

Y1 - 2021

N2 - The lack of standardized pipelines for image processing has prevented the application of deep learning (DL) techniques for the segmentation of the aorta in phase-contrast enhanced magnetic resonance angiography (PC-MRA). Furthermore, large, well-curated and annotated datasets, which are needed to create DL-based models able to generalize, are rare. We present the adaptation of the popular nnU-net DL framework to automatically segment the aorta in 4D flow MRI-derived angiograms. The resulting segmentations in a large database (> 300 cases) with normal cases and examples of different pathologies of the aorta provided from a single centre were excellent after post-processing (Dice score of 0.944). Subsequently, we explored the generalisation of the trained network in a small dataset of images (around 20 cases) acquired in a different hospital with another scanner. Without domain adaptation, only with a model trained with the large dataset, the obtained results were substantially worst than with adding a few cases of the small dataset (Dice scores of 0.61 vs 0.86, respectively). The obtained results created good quality segmentations of the aorta in 4D flow MRI, which can later be post-processed to assess blood flow patterns, similarly than with manual annotations. However, advanced domain adaptation schemes are very important in 4D flow MRI due to the large differences in image characteristics between different vendor scanners available in multiple centers.

AB - The lack of standardized pipelines for image processing has prevented the application of deep learning (DL) techniques for the segmentation of the aorta in phase-contrast enhanced magnetic resonance angiography (PC-MRA). Furthermore, large, well-curated and annotated datasets, which are needed to create DL-based models able to generalize, are rare. We present the adaptation of the popular nnU-net DL framework to automatically segment the aorta in 4D flow MRI-derived angiograms. The resulting segmentations in a large database (> 300 cases) with normal cases and examples of different pathologies of the aorta provided from a single centre were excellent after post-processing (Dice score of 0.944). Subsequently, we explored the generalisation of the trained network in a small dataset of images (around 20 cases) acquired in a different hospital with another scanner. Without domain adaptation, only with a model trained with the large dataset, the obtained results were substantially worst than with adding a few cases of the small dataset (Dice scores of 0.61 vs 0.86, respectively). The obtained results created good quality segmentations of the aorta in 4D flow MRI, which can later be post-processed to assess blood flow patterns, similarly than with manual annotations. However, advanced domain adaptation schemes are very important in 4D flow MRI due to the large differences in image characteristics between different vendor scanners available in multiple centers.

KW - 4D flow magnetic resonance imaging

KW - Aortic segmentation

KW - Deep learning

KW - nnU-net

UR - http://www.scopus.com/inward/record.url?scp=85111847835&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-78710-3_12

DO - 10.1007/978-3-030-78710-3_12

M3 - Conference paper

AN - SCOPUS:85111847835

SN - 9783030787097

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

SP - 112

EP - 121

BT - Functional Imaging and Modeling of the Heart - 11th International Conference, FIMH 2021, Proceedings

A2 - Ennis, Daniel B.

A2 - Perotti, Luigi E.

A2 - Wang, Vicky Y.

PB - Springer Science and Business Media Deutschland GmbH

T2 - 11th International Conference on Functional Imaging and Modeling of the Heart, FIMH 2021

Y2 - 21 June 2021 through 25 June 2021

ER -

Aviles J, Talou GDM, Camara O, Córdova MM, Ferez XM, Romero D et al. Domain Adaptation for Automatic Aorta Segmentation of 4D Flow Magnetic Resonance Imaging Data from Multiple Vendor Scanners. In Ennis DB, Perotti LE, Wang VY, editors, Functional Imaging and Modeling of the Heart - 11th International Conference, FIMH 2021, Proceedings. Springer Science and Business Media Deutschland GmbH. 2021. p. 112-121. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-78710-3_12

Domain Adaptation for Automatic Aorta Segmentation of 4D Flow Magnetic Resonance Imaging Data from Multiple Vendor Scanners

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this