Guiding 3D U-nets with signed distance fields for creating 3D models from images

Kristine Aavild Juhl; Rasmus Reinhold Paulsen; Anders Bjorholm Dahl; Vedrana Andersen Dahl; Ole De Backer; Klaus Fuglsang Kofoed; Oscar Camara

Guiding 3D U-nets with signed distance fields for creating 3D models from images

Kristine Aavild Juhl, Rasmus Reinhold Paulsen, Anders Bjorholm Dahl, Vedrana Andersen Dahl, Ole De Backer, Klaus Fuglsang Kofoed, Oscar Camara

Research output: Contribution to conference › Paper › Research › peer-review

183 Downloads (Pure)

Abstract

Morphological analysis of the left atrial appendage is an important tool to assess risk of ischemic stroke. Most deep learning approaches for 3D segmentation is guided by binary labelmaps, which results in voxelized segmentations unsuitable for morphological analysis. We propose to use signed distance fields to guide a deep network towards morphologically consistent 3D models. The proposed strategy is evaluated on a synthetic dataset of simple geometries, as well as a set of cardiac computed tomography images containing the left atrial appendage. The proposed method produces smooth surfaces with a closer resemblance to the true surface in terms of segmentation overlap and surface distance.

Original language	English
Publication date	2019
Number of pages	4
Publication status	Published - 2019
Event	Medical Imaging with Deep Learning 2019 - Imperial College’s South Kensington Campus, London, United Kingdom Duration: 8 Jul 2019 → 10 Jul 2019 http://2019.midl.io/

Conference

Conference	Medical Imaging with Deep Learning 2019
Location	Imperial College’s South Kensington Campus
Country/Territory	United Kingdom
City	London
Period	08/07/2019 → 10/07/2019
Internet address	http://2019.midl.io/

Bibliographical note

https://openreview.net/group?id=MIDL.io/2019/Conference/Abstract

Keywords

Signed distance fields
Pixel-wise regression
Left atrial appendage

Access to Document

FulltextFinal published version, 539 KB

OpenUrl availability

Full text

Cite this

@conference{282ac767181747d0a7e20db3580789c3,

title = "Guiding 3D U-nets with signed distance fields for creating 3D models from images",

abstract = "Morphological analysis of the left atrial appendage is an important tool to assess risk of ischemic stroke. Most deep learning approaches for 3D segmentation is guided by binary labelmaps, which results in voxelized segmentations unsuitable for morphological analysis. We propose to use signed distance fields to guide a deep network towards morphologically consistent 3D models. The proposed strategy is evaluated on a synthetic dataset of simple geometries, as well as a set of cardiac computed tomography images containing the left atrial appendage. The proposed method produces smooth surfaces with a closer resemblance to the true surface in terms of segmentation overlap and surface distance.",

keywords = "Signed distance fields, Pixel-wise regression, Left atrial appendage",

author = "Juhl, {Kristine Aavild} and Paulsen, {Rasmus Reinhold} and Dahl, {Anders Bjorholm} and Dahl, {Vedrana Andersen} and {De Backer}, Ole and Kofoed, {Klaus Fuglsang} and Oscar Camara",

note = "https://openreview.net/group?id=MIDL.io/2019/Conference/Abstract; Medical Imaging with Deep Learning 2019, MIDL 2019 ; Conference date: 08-07-2019 Through 10-07-2019",

year = "2019",

language = "English",

url = "http://2019.midl.io/",

}

TY - CONF

T1 - Guiding 3D U-nets with signed distance fields for creating 3D models from images

AU - Juhl, Kristine Aavild

AU - Paulsen, Rasmus Reinhold

AU - Dahl, Anders Bjorholm

AU - Dahl, Vedrana Andersen

AU - De Backer, Ole

AU - Kofoed, Klaus Fuglsang

AU - Camara, Oscar

N1 - https://openreview.net/group?id=MIDL.io/2019/Conference/Abstract

PY - 2019

Y1 - 2019

N2 - Morphological analysis of the left atrial appendage is an important tool to assess risk of ischemic stroke. Most deep learning approaches for 3D segmentation is guided by binary labelmaps, which results in voxelized segmentations unsuitable for morphological analysis. We propose to use signed distance fields to guide a deep network towards morphologically consistent 3D models. The proposed strategy is evaluated on a synthetic dataset of simple geometries, as well as a set of cardiac computed tomography images containing the left atrial appendage. The proposed method produces smooth surfaces with a closer resemblance to the true surface in terms of segmentation overlap and surface distance.

AB - Morphological analysis of the left atrial appendage is an important tool to assess risk of ischemic stroke. Most deep learning approaches for 3D segmentation is guided by binary labelmaps, which results in voxelized segmentations unsuitable for morphological analysis. We propose to use signed distance fields to guide a deep network towards morphologically consistent 3D models. The proposed strategy is evaluated on a synthetic dataset of simple geometries, as well as a set of cardiac computed tomography images containing the left atrial appendage. The proposed method produces smooth surfaces with a closer resemblance to the true surface in terms of segmentation overlap and surface distance.

KW - Signed distance fields

KW - Pixel-wise regression

KW - Left atrial appendage

M3 - Paper

T2 - Medical Imaging with Deep Learning 2019

Y2 - 8 July 2019 through 10 July 2019

ER -

Guiding 3D U-nets with signed distance fields for creating 3D models from images

Abstract

Conference

Bibliographical note

Keywords

Access to Document

OpenUrl availability

Fingerprint

Cite this