Breast pectoral muscle segmentation in mammograms using a modified holistically-nested edge detection network

Andrik Rampun, Karen López-Linares, Philip Morrow, Bryan Scotney, Hui Wang, Inmaculada Garcia Ocaña, Grégory Maclair, Reyer Zwiggelaar, Miguel A. González Ballester, Iván Macía

Research output: Contribution to journalArticle

Abstract

This paper presents a method for automatic breast pectoral muscle segmentation in mediolateral oblique mammograms using a Convolutional Neural Network (CNN) inspired by the Holistically-nested Edge Detection (HED) network. Most of the existing methods in the literature are based on hand-crafted models such as straight-line, curve-based techniques or a combination of both. Unfortunately, such models are insufficient when dealing with complex shape variations of the pectoral muscle boundary and when the boundary is unclear due to overlapping breast tissue. To compensate for these issues, we propose a neural network framework that incorporates multi-scale and multi-level learning, capable of learning complex hierarchical features to resolve spatial ambiguity in estimating the pectoral muscle boundary. For this purpose, we modified the HED network architecture to specifically find ‘contour-like’ objects in mammograms. The proposed framework produced a probability map that can be used to estimate the initial pectoral muscle boundary. Subsequently, we process these maps by extracting morphological properties to find the actual pectoral muscle boundary. Finally, we developed two different post-processing steps to find the actual pectoral muscle boundary. Quantitative evaluation results show that the proposed method is comparable with alternative state-of-the-art methods producing on average values of 94.8 ± 8.5% and 97.5 ± 6.3% for the Jaccard and Dice similarity metrics, respectively, across four different databases.
LanguageEnglish
Pages1-17
Number of pages17
JournalMedical Image Analysis
Volume57
Early online date20 Jun 2019
DOIs
Publication statusE-pub ahead of print - 20 Jun 2019

Fingerprint

Edge detection
Muscle
Neural networks
Network architecture
Tissue
Processing

Keywords

  • Breast mammography
  • Computer aided diagnosis
  • Convolutional neural networks
  • Deep learning
  • Pectoral muscle segmentation

Cite this

Rampun, Andrik ; López-Linares, Karen ; Morrow, Philip ; Scotney, Bryan ; Wang, Hui ; Garcia Ocaña, Inmaculada ; Maclair, Grégory ; Zwiggelaar, Reyer ; González Ballester, Miguel A. ; Macía, Iván. / Breast pectoral muscle segmentation in mammograms using a modified holistically-nested edge detection network. 2019 ; Vol. 57. pp. 1-17.
@article{e20a052944e74311896770ebb4ab3ba4,
title = "Breast pectoral muscle segmentation in mammograms using a modified holistically-nested edge detection network",
abstract = "This paper presents a method for automatic breast pectoral muscle segmentation in mediolateral oblique mammograms using a Convolutional Neural Network (CNN) inspired by the Holistically-nested Edge Detection (HED) network. Most of the existing methods in the literature are based on hand-crafted models such as straight-line, curve-based techniques or a combination of both. Unfortunately, such models are insufficient when dealing with complex shape variations of the pectoral muscle boundary and when the boundary is unclear due to overlapping breast tissue. To compensate for these issues, we propose a neural network framework that incorporates multi-scale and multi-level learning, capable of learning complex hierarchical features to resolve spatial ambiguity in estimating the pectoral muscle boundary. For this purpose, we modified the HED network architecture to specifically find ‘contour-like’ objects in mammograms. The proposed framework produced a probability map that can be used to estimate the initial pectoral muscle boundary. Subsequently, we process these maps by extracting morphological properties to find the actual pectoral muscle boundary. Finally, we developed two different post-processing steps to find the actual pectoral muscle boundary. Quantitative evaluation results show that the proposed method is comparable with alternative state-of-the-art methods producing on average values of 94.8 ± 8.5{\%} and 97.5 ± 6.3{\%} for the Jaccard and Dice similarity metrics, respectively, across four different databases.",
keywords = "Breast mammography, Computer aided diagnosis, Convolutional neural networks, Deep learning, Pectoral muscle segmentation",
author = "Andrik Rampun and Karen L{\'o}pez-Linares and Philip Morrow and Bryan Scotney and Hui Wang and {Garcia Oca{\~n}a}, Inmaculada and Gr{\'e}gory Maclair and Reyer Zwiggelaar and {Gonz{\'a}lez Ballester}, {Miguel A.} and Iv{\'a}n Mac{\'i}a",
year = "2019",
month = "6",
day = "20",
doi = "10.1016/j.media.2019.06.007",
language = "English",
volume = "57",
pages = "1--17",

}

Rampun, A, López-Linares, K, Morrow, P, Scotney, B, Wang, H, Garcia Ocaña, I, Maclair, G, Zwiggelaar, R, González Ballester, MA & Macía, I 2019, 'Breast pectoral muscle segmentation in mammograms using a modified holistically-nested edge detection network', vol. 57, pp. 1-17. https://doi.org/10.1016/j.media.2019.06.007

Breast pectoral muscle segmentation in mammograms using a modified holistically-nested edge detection network. / Rampun, Andrik; López-Linares, Karen; Morrow, Philip; Scotney, Bryan; Wang, Hui; Garcia Ocaña, Inmaculada; Maclair, Grégory; Zwiggelaar, Reyer; González Ballester, Miguel A. ; Macía, Iván.

Vol. 57, 31.10.2019, p. 1-17.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Breast pectoral muscle segmentation in mammograms using a modified holistically-nested edge detection network

AU - Rampun, Andrik

AU - López-Linares, Karen

AU - Morrow, Philip

AU - Scotney, Bryan

AU - Wang, Hui

AU - Garcia Ocaña, Inmaculada

AU - Maclair, Grégory

AU - Zwiggelaar, Reyer

AU - González Ballester, Miguel A.

AU - Macía, Iván

PY - 2019/6/20

Y1 - 2019/6/20

N2 - This paper presents a method for automatic breast pectoral muscle segmentation in mediolateral oblique mammograms using a Convolutional Neural Network (CNN) inspired by the Holistically-nested Edge Detection (HED) network. Most of the existing methods in the literature are based on hand-crafted models such as straight-line, curve-based techniques or a combination of both. Unfortunately, such models are insufficient when dealing with complex shape variations of the pectoral muscle boundary and when the boundary is unclear due to overlapping breast tissue. To compensate for these issues, we propose a neural network framework that incorporates multi-scale and multi-level learning, capable of learning complex hierarchical features to resolve spatial ambiguity in estimating the pectoral muscle boundary. For this purpose, we modified the HED network architecture to specifically find ‘contour-like’ objects in mammograms. The proposed framework produced a probability map that can be used to estimate the initial pectoral muscle boundary. Subsequently, we process these maps by extracting morphological properties to find the actual pectoral muscle boundary. Finally, we developed two different post-processing steps to find the actual pectoral muscle boundary. Quantitative evaluation results show that the proposed method is comparable with alternative state-of-the-art methods producing on average values of 94.8 ± 8.5% and 97.5 ± 6.3% for the Jaccard and Dice similarity metrics, respectively, across four different databases.

AB - This paper presents a method for automatic breast pectoral muscle segmentation in mediolateral oblique mammograms using a Convolutional Neural Network (CNN) inspired by the Holistically-nested Edge Detection (HED) network. Most of the existing methods in the literature are based on hand-crafted models such as straight-line, curve-based techniques or a combination of both. Unfortunately, such models are insufficient when dealing with complex shape variations of the pectoral muscle boundary and when the boundary is unclear due to overlapping breast tissue. To compensate for these issues, we propose a neural network framework that incorporates multi-scale and multi-level learning, capable of learning complex hierarchical features to resolve spatial ambiguity in estimating the pectoral muscle boundary. For this purpose, we modified the HED network architecture to specifically find ‘contour-like’ objects in mammograms. The proposed framework produced a probability map that can be used to estimate the initial pectoral muscle boundary. Subsequently, we process these maps by extracting morphological properties to find the actual pectoral muscle boundary. Finally, we developed two different post-processing steps to find the actual pectoral muscle boundary. Quantitative evaluation results show that the proposed method is comparable with alternative state-of-the-art methods producing on average values of 94.8 ± 8.5% and 97.5 ± 6.3% for the Jaccard and Dice similarity metrics, respectively, across four different databases.

KW - Breast mammography

KW - Computer aided diagnosis

KW - Convolutional neural networks

KW - Deep learning

KW - Pectoral muscle segmentation

UR - http://www.scopus.com/inward/record.url?scp=85067874853&partnerID=8YFLogxK

U2 - 10.1016/j.media.2019.06.007

DO - 10.1016/j.media.2019.06.007

M3 - Article

VL - 57

SP - 1

EP - 17

ER -