Automated Breast Segmentation of Fat and Water MR Images Using Dynamic Programming

José A. Rosado-Toro, Tomoe Barr, Jean-Philippe Galons, Marilyn T. Marron, Alison T Stopeck, Cynthia Thomson, Patricia Thompson, Danielle Carroll, Eszter Wolf, Maria I Altbach, Jeffrey J Rodriguez

Research output: Contribution to journalArticle

25 Citations (Scopus)

Abstract

Rationale and Objectives: To develop and test an algorithm that outlines the breast boundaries using information from fat and water magnetic resonance images. Materials and Methods: Three algorithms were implemented and tested using registered fat and water magnetic resonance images. Two of the segmentation algorithms are simple extensions of the techniques used for contrast-enhanced images: one algorithm uses clustering and local gradient (CLG) analysis and the other algorithm uses a Hessian-based sheetness filter (HSF). The third segmentation algorithm uses k-means++ and dynamic programming (KDP) for finding the breast pixels. All three algorithms separate the left and right breasts using either a fixed region or a morphological method. The performance is quantified using a mutual overlap (Dice) metric and a pectoral muscle boundary error. The algorithms are evaluated against three manual tracers using 266 breast images from 14 female subjects. Results: The KDP algorithm has a mean overlap percentage improvement that is statistically significant relative to the HSF and CLG algorithms. When using a fixed region to remove the tissue between breasts with tracer 1 as a reference, the KDP algorithm has a mean overlap of 0.922 compared to 0.864 (. P<.01) for HSF and 0.843 (. P<.01) for CLG. The performance of KDP is very similar to tracers 2 (0.926 overlap) and 3 (0.929 overlap). The performance analysis in terms of pectoral muscle boundary error showed that the fraction of the muscle boundary within three pixels of reference tracer 1 is 0.87 using KDP compared to 0.578 for HSF and 0.617 for CLG. Our results show that the performance of the KDP algorithm is independent of breast density. Conclusions: We developed a new automated segmentation algorithm (KDP) to isolate breast tissue from magnetic resonance fat and water images. KDP outperforms the other techniques that focus on local analysis (CLG and HSF) and yields a performance similar to human tracers.

Original languageEnglish (US)
Pages (from-to)139-148
Number of pages10
JournalAcademic Radiology
Volume22
Issue number2
DOIs
StatePublished - Feb 1 2015

Fingerprint

Breast
Fats
Water
Cluster Analysis
Pectoralis Muscles
Magnetic Resonance Spectroscopy
Muscles

Keywords

  • Automated breast segmentation
  • Breast MRI
  • Dynamic programming
  • Fat-water MRI
  • K-means++

ASJC Scopus subject areas

  • Radiology Nuclear Medicine and imaging

Cite this

Automated Breast Segmentation of Fat and Water MR Images Using Dynamic Programming. / Rosado-Toro, José A.; Barr, Tomoe; Galons, Jean-Philippe; Marron, Marilyn T.; Stopeck, Alison T; Thomson, Cynthia; Thompson, Patricia; Carroll, Danielle; Wolf, Eszter; Altbach, Maria I; Rodriguez, Jeffrey J.

In: Academic Radiology, Vol. 22, No. 2, 01.02.2015, p. 139-148.

Research output: Contribution to journalArticle

Rosado-Toro, José A. ; Barr, Tomoe ; Galons, Jean-Philippe ; Marron, Marilyn T. ; Stopeck, Alison T ; Thomson, Cynthia ; Thompson, Patricia ; Carroll, Danielle ; Wolf, Eszter ; Altbach, Maria I ; Rodriguez, Jeffrey J. / Automated Breast Segmentation of Fat and Water MR Images Using Dynamic Programming. In: Academic Radiology. 2015 ; Vol. 22, No. 2. pp. 139-148.
@article{d09c3c657c5e426fa0c65a8385971b93,
title = "Automated Breast Segmentation of Fat and Water MR Images Using Dynamic Programming",
abstract = "Rationale and Objectives: To develop and test an algorithm that outlines the breast boundaries using information from fat and water magnetic resonance images. Materials and Methods: Three algorithms were implemented and tested using registered fat and water magnetic resonance images. Two of the segmentation algorithms are simple extensions of the techniques used for contrast-enhanced images: one algorithm uses clustering and local gradient (CLG) analysis and the other algorithm uses a Hessian-based sheetness filter (HSF). The third segmentation algorithm uses k-means++ and dynamic programming (KDP) for finding the breast pixels. All three algorithms separate the left and right breasts using either a fixed region or a morphological method. The performance is quantified using a mutual overlap (Dice) metric and a pectoral muscle boundary error. The algorithms are evaluated against three manual tracers using 266 breast images from 14 female subjects. Results: The KDP algorithm has a mean overlap percentage improvement that is statistically significant relative to the HSF and CLG algorithms. When using a fixed region to remove the tissue between breasts with tracer 1 as a reference, the KDP algorithm has a mean overlap of 0.922 compared to 0.864 (. P<.01) for HSF and 0.843 (. P<.01) for CLG. The performance of KDP is very similar to tracers 2 (0.926 overlap) and 3 (0.929 overlap). The performance analysis in terms of pectoral muscle boundary error showed that the fraction of the muscle boundary within three pixels of reference tracer 1 is 0.87 using KDP compared to 0.578 for HSF and 0.617 for CLG. Our results show that the performance of the KDP algorithm is independent of breast density. Conclusions: We developed a new automated segmentation algorithm (KDP) to isolate breast tissue from magnetic resonance fat and water images. KDP outperforms the other techniques that focus on local analysis (CLG and HSF) and yields a performance similar to human tracers.",
keywords = "Automated breast segmentation, Breast MRI, Dynamic programming, Fat-water MRI, K-means++",
author = "Rosado-Toro, {Jos{\'e} A.} and Tomoe Barr and Jean-Philippe Galons and Marron, {Marilyn T.} and Stopeck, {Alison T} and Cynthia Thomson and Patricia Thompson and Danielle Carroll and Eszter Wolf and Altbach, {Maria I} and Rodriguez, {Jeffrey J}",
year = "2015",
month = "2",
day = "1",
doi = "10.1016/j.acra.2014.09.015",
language = "English (US)",
volume = "22",
pages = "139--148",
journal = "Academic Radiology",
issn = "1076-6332",
publisher = "Elsevier USA",
number = "2",

}

TY - JOUR

T1 - Automated Breast Segmentation of Fat and Water MR Images Using Dynamic Programming

AU - Rosado-Toro, José A.

AU - Barr, Tomoe

AU - Galons, Jean-Philippe

AU - Marron, Marilyn T.

AU - Stopeck, Alison T

AU - Thomson, Cynthia

AU - Thompson, Patricia

AU - Carroll, Danielle

AU - Wolf, Eszter

AU - Altbach, Maria I

AU - Rodriguez, Jeffrey J

PY - 2015/2/1

Y1 - 2015/2/1

N2 - Rationale and Objectives: To develop and test an algorithm that outlines the breast boundaries using information from fat and water magnetic resonance images. Materials and Methods: Three algorithms were implemented and tested using registered fat and water magnetic resonance images. Two of the segmentation algorithms are simple extensions of the techniques used for contrast-enhanced images: one algorithm uses clustering and local gradient (CLG) analysis and the other algorithm uses a Hessian-based sheetness filter (HSF). The third segmentation algorithm uses k-means++ and dynamic programming (KDP) for finding the breast pixels. All three algorithms separate the left and right breasts using either a fixed region or a morphological method. The performance is quantified using a mutual overlap (Dice) metric and a pectoral muscle boundary error. The algorithms are evaluated against three manual tracers using 266 breast images from 14 female subjects. Results: The KDP algorithm has a mean overlap percentage improvement that is statistically significant relative to the HSF and CLG algorithms. When using a fixed region to remove the tissue between breasts with tracer 1 as a reference, the KDP algorithm has a mean overlap of 0.922 compared to 0.864 (. P<.01) for HSF and 0.843 (. P<.01) for CLG. The performance of KDP is very similar to tracers 2 (0.926 overlap) and 3 (0.929 overlap). The performance analysis in terms of pectoral muscle boundary error showed that the fraction of the muscle boundary within three pixels of reference tracer 1 is 0.87 using KDP compared to 0.578 for HSF and 0.617 for CLG. Our results show that the performance of the KDP algorithm is independent of breast density. Conclusions: We developed a new automated segmentation algorithm (KDP) to isolate breast tissue from magnetic resonance fat and water images. KDP outperforms the other techniques that focus on local analysis (CLG and HSF) and yields a performance similar to human tracers.

AB - Rationale and Objectives: To develop and test an algorithm that outlines the breast boundaries using information from fat and water magnetic resonance images. Materials and Methods: Three algorithms were implemented and tested using registered fat and water magnetic resonance images. Two of the segmentation algorithms are simple extensions of the techniques used for contrast-enhanced images: one algorithm uses clustering and local gradient (CLG) analysis and the other algorithm uses a Hessian-based sheetness filter (HSF). The third segmentation algorithm uses k-means++ and dynamic programming (KDP) for finding the breast pixels. All three algorithms separate the left and right breasts using either a fixed region or a morphological method. The performance is quantified using a mutual overlap (Dice) metric and a pectoral muscle boundary error. The algorithms are evaluated against three manual tracers using 266 breast images from 14 female subjects. Results: The KDP algorithm has a mean overlap percentage improvement that is statistically significant relative to the HSF and CLG algorithms. When using a fixed region to remove the tissue between breasts with tracer 1 as a reference, the KDP algorithm has a mean overlap of 0.922 compared to 0.864 (. P<.01) for HSF and 0.843 (. P<.01) for CLG. The performance of KDP is very similar to tracers 2 (0.926 overlap) and 3 (0.929 overlap). The performance analysis in terms of pectoral muscle boundary error showed that the fraction of the muscle boundary within three pixels of reference tracer 1 is 0.87 using KDP compared to 0.578 for HSF and 0.617 for CLG. Our results show that the performance of the KDP algorithm is independent of breast density. Conclusions: We developed a new automated segmentation algorithm (KDP) to isolate breast tissue from magnetic resonance fat and water images. KDP outperforms the other techniques that focus on local analysis (CLG and HSF) and yields a performance similar to human tracers.

KW - Automated breast segmentation

KW - Breast MRI

KW - Dynamic programming

KW - Fat-water MRI

KW - K-means++

UR - http://www.scopus.com/inward/record.url?scp=84920281243&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84920281243&partnerID=8YFLogxK

U2 - 10.1016/j.acra.2014.09.015

DO - 10.1016/j.acra.2014.09.015

M3 - Article

C2 - 25572926

AN - SCOPUS:84920281243

VL - 22

SP - 139

EP - 148

JO - Academic Radiology

JF - Academic Radiology

SN - 1076-6332

IS - 2

ER -