Automatic Vessel Lumen Segmentation in Optical Coherence Tomograph (OCT) Images

HUAIZHONG ZHANG, Ehab Essa, Xianghua Xie

Research output: Contribution to journalArticle

Abstract

This paper presents a graph based method to automatically and accurately segment the lumen borders from optical coherence tomography (OCT) images.
The proposed method unravels the OCT images from the Cartesian coordinates to polar coordinates so that the segmentation is transferred into a height field delineation problem. In effect, the method imposes a simplistic star shape prior but without the bias towards narrower lumen size. The lumen border is identified as the solution to finding the minimum closed set on a node weighed, directed graph. In order to cope with both the variability in imaging condition and different forms of image artefacts, we adopt an image feature that relies on very little assumption on the appearance of lumen border but is resilient to image noise and so on. This feature is derived from a convolution of the image gradient field and thus it takes into account gradient vector interactions at a much more global scale compared to conventional gradient based approaches. The proposed method is fully automatic without the need for an initialisation. We compare this method with a number of techniques, including both conventional methods and data driven models.
Original languageEnglish
Pages (from-to)1-11
JournalApplied Soft Computing Journal
Volume88
Early online date2 Jan 2020
DOIs
Publication statusE-pub ahead of print - 2 Jan 2020

Fingerprint

Optical tomography
Directed graphs
Convolution
Stars
Imaging techniques

Keywords

  • Image segmentation
  • Gradient convolution field
  • Star prior
  • Optical Coherence Tomography
  • Graph Cuts

Cite this

@article{e78b556640804576ac8a1571346db5f7,
title = "Automatic Vessel Lumen Segmentation in Optical Coherence Tomograph (OCT) Images",
abstract = "This paper presents a graph based method to automatically and accurately segment the lumen borders from optical coherence tomography (OCT) images.The proposed method unravels the OCT images from the Cartesian coordinates to polar coordinates so that the segmentation is transferred into a height field delineation problem. In effect, the method imposes a simplistic star shape prior but without the bias towards narrower lumen size. The lumen border is identified as the solution to finding the minimum closed set on a node weighed, directed graph. In order to cope with both the variability in imaging condition and different forms of image artefacts, we adopt an image feature that relies on very little assumption on the appearance of lumen border but is resilient to image noise and so on. This feature is derived from a convolution of the image gradient field and thus it takes into account gradient vector interactions at a much more global scale compared to conventional gradient based approaches. The proposed method is fully automatic without the need for an initialisation. We compare this method with a number of techniques, including both conventional methods and data driven models.",
keywords = "Image segmentation, Gradient convolution field, Star prior, Optical Coherence Tomography, Graph Cuts",
author = "HUAIZHONG ZHANG and Ehab Essa and Xianghua Xie",
year = "2020",
month = "1",
day = "2",
doi = "10.1016/j.asoc.2019.106042",
language = "English",
volume = "88",
pages = "1--11",
journal = "Applied Soft Computing Journal",
issn = "1568-4946",
publisher = "Elsevier",

}

Automatic Vessel Lumen Segmentation in Optical Coherence Tomograph (OCT) Images. / ZHANG, HUAIZHONG; Essa, Ehab; Xie, Xianghua.

In: Applied Soft Computing Journal, Vol. 88, 02.01.2020, p. 1-11.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Automatic Vessel Lumen Segmentation in Optical Coherence Tomograph (OCT) Images

AU - ZHANG, HUAIZHONG

AU - Essa, Ehab

AU - Xie, Xianghua

PY - 2020/1/2

Y1 - 2020/1/2

N2 - This paper presents a graph based method to automatically and accurately segment the lumen borders from optical coherence tomography (OCT) images.The proposed method unravels the OCT images from the Cartesian coordinates to polar coordinates so that the segmentation is transferred into a height field delineation problem. In effect, the method imposes a simplistic star shape prior but without the bias towards narrower lumen size. The lumen border is identified as the solution to finding the minimum closed set on a node weighed, directed graph. In order to cope with both the variability in imaging condition and different forms of image artefacts, we adopt an image feature that relies on very little assumption on the appearance of lumen border but is resilient to image noise and so on. This feature is derived from a convolution of the image gradient field and thus it takes into account gradient vector interactions at a much more global scale compared to conventional gradient based approaches. The proposed method is fully automatic without the need for an initialisation. We compare this method with a number of techniques, including both conventional methods and data driven models.

AB - This paper presents a graph based method to automatically and accurately segment the lumen borders from optical coherence tomography (OCT) images.The proposed method unravels the OCT images from the Cartesian coordinates to polar coordinates so that the segmentation is transferred into a height field delineation problem. In effect, the method imposes a simplistic star shape prior but without the bias towards narrower lumen size. The lumen border is identified as the solution to finding the minimum closed set on a node weighed, directed graph. In order to cope with both the variability in imaging condition and different forms of image artefacts, we adopt an image feature that relies on very little assumption on the appearance of lumen border but is resilient to image noise and so on. This feature is derived from a convolution of the image gradient field and thus it takes into account gradient vector interactions at a much more global scale compared to conventional gradient based approaches. The proposed method is fully automatic without the need for an initialisation. We compare this method with a number of techniques, including both conventional methods and data driven models.

KW - Image segmentation

KW - Gradient convolution field

KW - Star prior

KW - Optical Coherence Tomography

KW - Graph Cuts

U2 - 10.1016/j.asoc.2019.106042

DO - 10.1016/j.asoc.2019.106042

M3 - Article

VL - 88

SP - 1

EP - 11

JO - Applied Soft Computing Journal

JF - Applied Soft Computing Journal

SN - 1568-4946

ER -