26-06-2012, 04:26 PM
Detection of Architectural Distortion in Mammograms Acquired Prior to the Detection of Breast Cancer using Texture and Fractal Analysis
[attachment=25631]
1. INTRODUCTION
Architectural distortion is defined in BI-RADS1 as follows: “The normal architecture (of the breast) is distorted
with no definite mass visible. This includes spiculations radiating from a point and focal retraction or distortion
at the edge of the parenchyma. Architectural distortion can also be an associated finding.” Focal retraction
is considered to be easier to perceive than spiculated distortion within the breast parenchyma. Architectural
distortion could be categorized as malignant or benign, the former including cancer and the latter including scar
and soft-tissue damage due to trauma.
Architectural distortion is the third most common mammographic sign of non-palpable breast cancer,2 but due
to its subtlety and variable presentation, it is often missed during screening. Specifically, architectural distortion
accounts for 12% to 45% of breast cancers overlooked or misinterpreted in screening mammography.3, 4 The
extent of errors due to overlooking of lesions reinforces the need for computer-aided diagnosis (CAD) tools in
mammography.5, 6 Various CAD techniques and systems have been proposed to enhance the sensitivity of the
detection of breast cancer. Although these techniques are effective in detecting masses and calcifications, they
have failed in detecting architectural distortion with a high level of accuracy.7
Computer-aided Diagnosis of Architectural Distortion
Burrrell et al.,8 in a study of screening interval breast cancers, showed that architectural distortion is the most
commonly missed abnormality in false-negative cases. Broeders et al.9 suggested that improvement in the
detection of architectural distortion could lead to an effective improvement in the prognosis of breast cancer
patients.
The presence of architectural distortion changes the normal oriented texture of the breast. Methods for the
characterization and detection of such subtle changes from a pattern recognition perspective were proposed by
Ayres and Rangayyan,10 including Gabor filters and phase portrait maps. The methods were tested with one set
of 19 cases of architectural distortion and 41 normal mammograms, and another set of 37 cases of architectural
distortion. The resulting free-response receiver operating characteristics (FROC) curve gave the sensitivity rates
of 84% at 4.5 false positives per image and 81% at 10 false positives per image for the two sets of images.10
Detection of Potential Sites of Architectural Distortion
We use the methods proposed by Ayres and Rangayyan10 for the detection of architectural distortion in mammograms,
based on the analysis of oriented texture through the application of Gabor filters and a linear phase
portrait model. The images are first filtered and downsampled to 200 μm/pixel. The breast portion of a given
mammogram is segmented by applying Otsu’s threshold,24 followed by the application of the morphological
opening filter25 with a disk-shaped structuring element of radius 25 pixels (5 mm at 200 μm per pixel) to
smooth the resulting edges. The method for the detection of architectural distortion consists of the following
stages: extraction of the orientation field using Gabor filters, selection of curvilinear structures (CLS), filtering
and downsampling of the orientation field, modeling of phase portraits, and detection of sites of architectural
distortion.
Estimation of Fractal Dimension
In order to estimate FD, the two-dimensional (2D) Fourier power spectrum of the ROI being processed is
obtained, including the application of the von Hann (also known as Hanning) window and zero padding to twice
the original size. The 2D power spectrum is transformed into a one-dimensional (1D) function by circularly
averaging as a function of the radial distance from the zero-frequency point. The resulting 1D power spectrum
P(f) represents the average value in the 2D power spectrum, for a given radial distance from the origin, over all
angles. The spectrum P(f) is considered to be related to the frequency f according to the model
MAMMOGRAPHIC DATA ACQUISITION
Mammographic images were selected from a database of 1,745 digitized mammograms of 170 subjects obtained
from Screen Test: Alberta Program for the Early Detection of Breast Cancer.29, 30 Ethical approval of the project
was obtained from the Conjoint Health Research Ethics Board, Office of Medical Bioethics, University of Calgary,
and the Calgary Regional Health Authority. All cases of screen-detected cancer with prior mammograms available
were included in the dataset for the present study, with the diagnosis confirmed by biopsy.
[attachment=25631]
1. INTRODUCTION
Architectural distortion is defined in BI-RADS1 as follows: “The normal architecture (of the breast) is distorted
with no definite mass visible. This includes spiculations radiating from a point and focal retraction or distortion
at the edge of the parenchyma. Architectural distortion can also be an associated finding.” Focal retraction
is considered to be easier to perceive than spiculated distortion within the breast parenchyma. Architectural
distortion could be categorized as malignant or benign, the former including cancer and the latter including scar
and soft-tissue damage due to trauma.
Architectural distortion is the third most common mammographic sign of non-palpable breast cancer,2 but due
to its subtlety and variable presentation, it is often missed during screening. Specifically, architectural distortion
accounts for 12% to 45% of breast cancers overlooked or misinterpreted in screening mammography.3, 4 The
extent of errors due to overlooking of lesions reinforces the need for computer-aided diagnosis (CAD) tools in
mammography.5, 6 Various CAD techniques and systems have been proposed to enhance the sensitivity of the
detection of breast cancer. Although these techniques are effective in detecting masses and calcifications, they
have failed in detecting architectural distortion with a high level of accuracy.7
Computer-aided Diagnosis of Architectural Distortion
Burrrell et al.,8 in a study of screening interval breast cancers, showed that architectural distortion is the most
commonly missed abnormality in false-negative cases. Broeders et al.9 suggested that improvement in the
detection of architectural distortion could lead to an effective improvement in the prognosis of breast cancer
patients.
The presence of architectural distortion changes the normal oriented texture of the breast. Methods for the
characterization and detection of such subtle changes from a pattern recognition perspective were proposed by
Ayres and Rangayyan,10 including Gabor filters and phase portrait maps. The methods were tested with one set
of 19 cases of architectural distortion and 41 normal mammograms, and another set of 37 cases of architectural
distortion. The resulting free-response receiver operating characteristics (FROC) curve gave the sensitivity rates
of 84% at 4.5 false positives per image and 81% at 10 false positives per image for the two sets of images.10
Detection of Potential Sites of Architectural Distortion
We use the methods proposed by Ayres and Rangayyan10 for the detection of architectural distortion in mammograms,
based on the analysis of oriented texture through the application of Gabor filters and a linear phase
portrait model. The images are first filtered and downsampled to 200 μm/pixel. The breast portion of a given
mammogram is segmented by applying Otsu’s threshold,24 followed by the application of the morphological
opening filter25 with a disk-shaped structuring element of radius 25 pixels (5 mm at 200 μm per pixel) to
smooth the resulting edges. The method for the detection of architectural distortion consists of the following
stages: extraction of the orientation field using Gabor filters, selection of curvilinear structures (CLS), filtering
and downsampling of the orientation field, modeling of phase portraits, and detection of sites of architectural
distortion.
Estimation of Fractal Dimension
In order to estimate FD, the two-dimensional (2D) Fourier power spectrum of the ROI being processed is
obtained, including the application of the von Hann (also known as Hanning) window and zero padding to twice
the original size. The 2D power spectrum is transformed into a one-dimensional (1D) function by circularly
averaging as a function of the radial distance from the zero-frequency point. The resulting 1D power spectrum
P(f) represents the average value in the 2D power spectrum, for a given radial distance from the origin, over all
angles. The spectrum P(f) is considered to be related to the frequency f according to the model
MAMMOGRAPHIC DATA ACQUISITION
Mammographic images were selected from a database of 1,745 digitized mammograms of 170 subjects obtained
from Screen Test: Alberta Program for the Early Detection of Breast Cancer.29, 30 Ethical approval of the project
was obtained from the Conjoint Health Research Ethics Board, Office of Medical Bioethics, University of Calgary,
and the Calgary Regional Health Authority. All cases of screen-detected cancer with prior mammograms available
were included in the dataset for the present study, with the diagnosis confirmed by biopsy.