Semi-supervised Virtual Regression of Aortic Dissections Using 3D Generative Inpainting

Antonio Pepe; Gabriel Mistelbauer; Christina Schwarz-Gsaxner; Jianning Li; D. Fleischmann; Dieter Schmalstieg; Jan Egger

doi:10.1007/978-3-030-62469-9_12

Semi-supervised Virtual Regression of Aortic Dissections Using 3D Generative Inpainting

Antonio Pepe^*
, Gabriel Mistelbauer
, Christina Schwarz-Gsaxner
, Jianning Li
, D. Fleischmann
, Dieter Schmalstieg
, Jan Egger^*

^*Corresponding author for this work

Institute of Visual Computing (7100)

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

Abstract

Aortic dissection (AD) is a condition of the main artery of the human body, resulting in the formation of a new flow channel, or false lumen (FL). The disease is usually diagnosed with a computed tomography angiography (CTA) scan during the acute phase. A better understanding of the causes of AD requires knowledge of aortic geometry prior to the event, which is available only in very rare circumstances. In this work, we propose an approach to reconstruct the aorta before the formation of a dissection by performing 3D inpainting with a two-stage generative adversarial network (GAN). In the first stage of our two-stage GAN, a network is trained on the 3D edge information of the healthy aorta to reconstruct the aortic wall. The second stage infers the image information of the aorta to reconstruct the entire dataset. We train our two-stage GAN with 3D patches from 55 non-dissected aortic datasets and evaluate it on 20 more non-dissected datasets, demonstrating that our proposed 3D architecture outperforms its 2D counterpart. To obtain pre-dissection aortae, we mask the entire FL in AD datasets. Finally, we provide qualitative feedback from a renown expert on the obtained pre-dissection cases

Original language	English
Title of host publication	Thoracic Image Analysis - Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Proceedings
Subtitle of host publication	Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings
Editors	Jens Petersen, Raúl San José Estépar, Alexander Schmidt-Richberg, Sarah Gerard, Bianca Lassen-Schmidt, Colin Jacobs, Reinhard Beichel, Kensaku Mori
Place of Publication	Cham
Publisher	Springer
Pages	130-140
Number of pages	11
ISBN (Electronic)	978-3-030-62469-9
ISBN (Print)	978-3-030-62468-2
DOIs	https://doi.org/10.1007/978-3-030-62469-9_12
Publication status	E-pub ahead of print - Nov 2020
Event	2nd International Workshop on Thoracic Image Analysis - Virtuell, Peru Duration: 8 Oct 2020 → 8 Oct 2020

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12502 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	2nd International Workshop on Thoracic Image Analysis
Abbreviated title	TIA 2020
Country/Territory	Peru
City	Virtuell
Period	8/10/20 → 8/10/20

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 3 Good Health and Well-being

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Keywords

Aortic dissection
Deep learning
Edge reconstruction
Generative adversarial networks
Inpainting

Access to Document

10.1007/978-3-030-62469-9_12

Cite this

Pepe, A., Mistelbauer, G., Schwarz-Gsaxner, C., Li, J., Fleischmann, D., Schmalstieg, D., & Egger, J. (2020). Semi-supervised Virtual Regression of Aortic Dissections Using 3D Generative Inpainting. In J. Petersen, R. San José Estépar, A. Schmidt-Richberg, S. Gerard, B. Lassen-Schmidt, C. Jacobs, R. Beichel, & K. Mori (Eds.), Thoracic Image Analysis - Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Proceedings: Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings (pp. 130-140). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12502 LNCS). Springer. Advance online publication. https://doi.org/10.1007/978-3-030-62469-9_12

Semi-supervised Virtual Regression of Aortic Dissections Using 3D Generative Inpainting. / Pepe, Antonio; Mistelbauer, Gabriel; Schwarz-Gsaxner, Christina et al.
Thoracic Image Analysis - Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Proceedings: Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings. ed. / Jens Petersen; Raúl San José Estépar; Alexander Schmidt-Richberg; Sarah Gerard; Bianca Lassen-Schmidt; Colin Jacobs; Reinhard Beichel; Kensaku Mori. Cham: Springer, 2020. p. 130-140 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12502 LNCS).

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

Pepe, A, Mistelbauer, G, Schwarz-Gsaxner, C, Li, J, Fleischmann, D, Schmalstieg, D & Egger, J 2020, Semi-supervised Virtual Regression of Aortic Dissections Using 3D Generative Inpainting. in J Petersen, R San José Estépar, A Schmidt-Richberg, S Gerard, B Lassen-Schmidt, C Jacobs, R Beichel & K Mori (eds), Thoracic Image Analysis - Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Proceedings: Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12502 LNCS, Springer, Cham, pp. 130-140, 2nd International Workshop on Thoracic Image Analysis, Virtuell, Peru, 8/10/20. https://doi.org/10.1007/978-3-030-62469-9_12

Pepe A, Mistelbauer G, Schwarz-Gsaxner C, Li J, Fleischmann D, Schmalstieg D et al. Semi-supervised Virtual Regression of Aortic Dissections Using 3D Generative Inpainting. In Petersen J, San José Estépar R, Schmidt-Richberg A, Gerard S, Lassen-Schmidt B, Jacobs C, Beichel R, Mori K, editors, Thoracic Image Analysis - Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Proceedings: Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings. Cham: Springer. 2020. p. 130-140. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). Epub 2020 Nov. doi: 10.1007/978-3-030-62469-9_12

Pepe, Antonio ; Mistelbauer, Gabriel ; Schwarz-Gsaxner, Christina et al. / Semi-supervised Virtual Regression of Aortic Dissections Using 3D Generative Inpainting. Thoracic Image Analysis - Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Proceedings: Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Lima, Peru, October 8, 2020, Proceedings. editor / Jens Petersen ; Raúl San José Estépar ; Alexander Schmidt-Richberg ; Sarah Gerard ; Bianca Lassen-Schmidt ; Colin Jacobs ; Reinhard Beichel ; Kensaku Mori. Cham : Springer, 2020. pp. 130-140 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{ee86af8a1e124328aff2a9df6e89f1ab,

title = "Semi-supervised Virtual Regression of Aortic Dissections Using 3D Generative Inpainting",

abstract = "Aortic dissection (AD) is a condition of the main artery of the human body, resulting in the formation of a new flow channel, or false lumen (FL). The disease is usually diagnosed with a computed tomography angiography (CTA) scan during the acute phase. A better understanding of the causes of AD requires knowledge of aortic geometry prior to the event, which is available only in very rare circumstances. In this work, we propose an approach to reconstruct the aorta before the formation of a dissection by performing 3D inpainting with a two-stage generative adversarial network (GAN). In the first stage of our two-stage GAN, a network is trained on the 3D edge information of the healthy aorta to reconstruct the aortic wall. The second stage infers the image information of the aorta to reconstruct the entire dataset. We train our two-stage GAN with 3D patches from 55 non-dissected aortic datasets and evaluate it on 20 more non-dissected datasets, demonstrating that our proposed 3D architecture outperforms its 2D counterpart. To obtain pre-dissection aortae, we mask the entire FL in AD datasets. Finally, we provide qualitative feedback from a renown expert on the obtained pre-dissection cases",

keywords = "Aortic dissection, Deep learning, Edge reconstruction, Generative adversarial networks, Inpainting",

author = "Antonio Pepe and Gabriel Mistelbauer and Christina Schwarz-Gsaxner and Jianning Li and D. Fleischmann and Dieter Schmalstieg and Jan Egger",

year = "2020",

month = nov,

doi = "10.1007/978-3-030-62469-9\_12",

language = "English",

isbn = "978-3-030-62468-2",

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

publisher = "Springer",

pages = "130--140",

editor = "Jens Petersen and \{San Jos{\'e} Est{\'e}par\}, Ra{\'u}l and Alexander Schmidt-Richberg and Sarah Gerard and Bianca Lassen-Schmidt and Colin Jacobs and Reinhard Beichel and Kensaku Mori",

booktitle = "Thoracic Image Analysis - Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Proceedings",

address = "United States",

note = "2nd International Workshop on Thoracic Image Analysis, TIA 2020 ; Conference date: 08-10-2020 Through 08-10-2020",

}

TY - GEN

T1 - Semi-supervised Virtual Regression of Aortic Dissections Using 3D Generative Inpainting

AU - Pepe, Antonio

AU - Mistelbauer, Gabriel

AU - Schwarz-Gsaxner, Christina

AU - Li, Jianning

AU - Fleischmann, D.

AU - Schmalstieg, Dieter

AU - Egger, Jan

PY - 2020/11

Y1 - 2020/11

N2 - Aortic dissection (AD) is a condition of the main artery of the human body, resulting in the formation of a new flow channel, or false lumen (FL). The disease is usually diagnosed with a computed tomography angiography (CTA) scan during the acute phase. A better understanding of the causes of AD requires knowledge of aortic geometry prior to the event, which is available only in very rare circumstances. In this work, we propose an approach to reconstruct the aorta before the formation of a dissection by performing 3D inpainting with a two-stage generative adversarial network (GAN). In the first stage of our two-stage GAN, a network is trained on the 3D edge information of the healthy aorta to reconstruct the aortic wall. The second stage infers the image information of the aorta to reconstruct the entire dataset. We train our two-stage GAN with 3D patches from 55 non-dissected aortic datasets and evaluate it on 20 more non-dissected datasets, demonstrating that our proposed 3D architecture outperforms its 2D counterpart. To obtain pre-dissection aortae, we mask the entire FL in AD datasets. Finally, we provide qualitative feedback from a renown expert on the obtained pre-dissection cases

AB - Aortic dissection (AD) is a condition of the main artery of the human body, resulting in the formation of a new flow channel, or false lumen (FL). The disease is usually diagnosed with a computed tomography angiography (CTA) scan during the acute phase. A better understanding of the causes of AD requires knowledge of aortic geometry prior to the event, which is available only in very rare circumstances. In this work, we propose an approach to reconstruct the aorta before the formation of a dissection by performing 3D inpainting with a two-stage generative adversarial network (GAN). In the first stage of our two-stage GAN, a network is trained on the 3D edge information of the healthy aorta to reconstruct the aortic wall. The second stage infers the image information of the aorta to reconstruct the entire dataset. We train our two-stage GAN with 3D patches from 55 non-dissected aortic datasets and evaluate it on 20 more non-dissected datasets, demonstrating that our proposed 3D architecture outperforms its 2D counterpart. To obtain pre-dissection aortae, we mask the entire FL in AD datasets. Finally, we provide qualitative feedback from a renown expert on the obtained pre-dissection cases

KW - Aortic dissection

KW - Deep learning

KW - Edge reconstruction

KW - Generative adversarial networks

KW - Inpainting

UR - https://www.scopus.com/pages/publications/85097268344

U2 - 10.1007/978-3-030-62469-9_12

DO - 10.1007/978-3-030-62469-9_12

M3 - Conference paper

SN - 978-3-030-62468-2

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

SP - 130

EP - 140

BT - Thoracic Image Analysis - Second International Workshop, TIA 2020, Held in Conjunction with MICCAI 2020, Proceedings

A2 - Petersen, Jens

A2 - San José Estépar, Raúl

A2 - Schmidt-Richberg, Alexander

A2 - Gerard, Sarah

A2 - Lassen-Schmidt, Bianca

A2 - Jacobs, Colin

A2 - Beichel, Reinhard

A2 - Mori, Kensaku

PB - Springer

CY - Cham

T2 - 2nd International Workshop on Thoracic Image Analysis

Y2 - 8 October 2020 through 8 October 2020

ER -

Semi-supervised Virtual Regression of Aortic Dissections Using 3D Generative Inpainting

Abstract

Publication series

Conference

UN SDGs

ASJC Scopus subject areas

Keywords

Access to Document

Other files and links

Fingerprint

Aortic Dissection