22-02-2013, 04:42 PM
Segmentation of the breast region in mammograms using active
contours
Segmentation.pdf (Size: 254.29 KB / Downloads: 46)
ABSTRACT
The largest single feature on a mammogram is the skin-air interface, or breast contour. Extraction of the breast contour is
useful for a number of reasons. Foremost it allows the search for abnormalities to be limited to the region of the breast
without undue influence from the background of the mammogram. Segmentation of the breast-region from the
background is made difficult by the tapering nature of the breast, such that the breast contour lies in between the softtissue
and the non-breast region. This paper explores the application of active contours to the problem of extracting the
breast region in mammograms.
INTRODUCTION
Breast cancer is the most frequently diagnosed form of cancer in Canadian women, accounting for approximately 30% of
all new cancer cases each year. In 2001, 19,500 cases were diagnosed and 5,500 women died of this disease1. In 1995,
breast cancer accounted for 97,000 potential years of life lost for Canadian women. Since 1984 breast cancer incidence
rates for all ages have increased progressively whilst mortality rates have remained relatively stable. One in nine
Canadian women is expected to develop breast cancer in her lifetime, and one out of every 25 is expected to die from it.
The role of the radiologist in breast cancer involves the interpretation of mammograms for the identification (detection)
of potential abnormalities and their classification (diagnosis) with respect to malignancy. Mammogram interpretation can
be hampered by the great variability in the appearance of normal and abnormal tissue within an image. As a result, the
effectiveness of mammography is limited by the radiologist's ability to accurately detect subtle, irregular abnormalities
embedded in the complex and highly textured nature of normal breast tissue. Although mammography is considered the
most reliable means of detecting breast cancer, between 10-30% of women subsequently diagnosed with breast cancer
have false-negative mammograms2. In approximately two-thirds of these false-negative mammograms, the radiologist
failed to detect a cancer that could be seen in retrospect on an earlier mammogram3.
SEGMENTATION OF THE BREAST REGION
A mammogram contains two distinctive regions: the exposed breast region and the unexposed air-background (nonbreast)
region. The principal feature on a mammogram is the breast contour, otherwise known as the skin-air interface,
or breast boundary. The breast contour can be obtained by partitioning the mammogram into breast and non-breast
regions. The extracted breast contour should adequately model the soft-tissue/air interface and preserve the nipple in
profile. The precise segmentation of the breast region in mammograms is an essential preprocessing step in the
computer-aided analysis of mammograms for a number of reasons. Foremost it allows the search for abnormalities to be
limited to the region of the breast without undue influence from the background of the mammogram. It also facilitates
enhancements to techniques such as comparative analysis, which includes the automated comparison of corresponding
mammograms. The breast boundary contains significant information relating to the deformation between two
mammograms and is the source of information for relating the position of the nipple relative to the skin surface.
However there are two major problems associated with the accurate segmentation of the breast region. Due to the
mammogram acquisition process, there is a region of decreasing contrast near the breast contour where the breast tapers
off. This region constitutes the uncompressed region of the breast commonly referred to as the "breast edge", and is
caused by a lack of uniform compression of the breast tissue. This tapering effect causes a lack of visibility along the
peripheral region of the mammogram, making it difficult to perceive the breast contour. Another problem is the nonuniformity
of the background region, which may contain high-intensity features such as information labels, artifacts
(e.g. scratches), or unexposed film regions.
Existing Approaches
Segmentation of the breast region in mammograms has traditionally been achieved using methods besides active
contours. One of the earliest approaches to segmentation of the breast contour was presented by Semmlow et al.9, who
used a spatial filter and Sobel edge detector to locate the breast boundary on xeromammograms. In many cases global
thresholding has been used to segment the breast region from the background10-11. The major problem with using global
thresholding is the nonuniform background region, although recent efforts, such as that of Masek et al.12 using local
thresholding have shown more promise. Abdel-Mottaleb et al.13 use a system of masking images with different
thresholds to find the breast edge. Gradients are calculated from the images as to where the skin should be located, and
the entire breast area can be found by a union of two images. Of the 500 mammograms tested, an "acceptable" boundary
was found in 98% of images. Méndez et al.14 find the breast contour using a gradient based method. They first use a twolevel
thresholding technique to isolate the breast region of the mammogram. The mammogram, oriented with the
mammogram facing upwards and smoothed, is then divided into three regions using a number of automatically
determined reference points and a tracking algorithm is applied to the mammogram to detect the border. The algorithm is
tested on 156 mammograms of which the breast contour is deemed to be "accurate" or "nearly accurate" in 89% of the
images. The global segmentation approach cited by Bick et al.15
Active Contours and Mammograms
There are several reasons why active contours offer an appropriate approach to the process of breast region extraction.
The principal rationale is that the breast is a well defined curve and as such is amenable to the curve approximation
characteristics of active contours. In addition, the background in most mammograms is a low intensity, low gradient
region, and as such can be avoided by the active contour in its search for a local minimum. However it is anticipated that
there will be several issues to resolve before using active contours with mammograms:
• Medium intensity noise may inadvertently attract the active contour away from the breast region if it is too
close to the initial contour.
• The breast-air interface is typically a medium gradient, so any energy functional based on edges will need some
preprocessing.
• The initial contour will have to be placed relatively close to the desired breast contour.