Anatomic breast coordinate system for mammogram analysis: 5th International Workshop on Breast Densitometry and Breast Cancer Risk Assessment ,San Francisco, United States

Gopal Karemore, S. Brandt, N. Karssemeijer, Mads Nielsen

Abstract

Purpose
Many researchers have investigated measures also other than density in the mammogram such as measures based on texture to improve breast cancer risk assessment. However, parenchymal texture characteristics are highly dependent on the orientation of vasculature structure and fibrous tissue inside the breast. Most of the risk assessment and CAD modules use a breast region in a image centered Cartesian x,y coordinate system. Nevertheless, anatomical structure follows curve-linear trajectories. We examined an anatomical breast coordinate system that preserves the anatomical correspondence between the mammograms and allows extracting not only the aligned position but also the orientation aligned with the anatomy of the breast tissue structure.

Materials and Methods
The coordinate system used the nipple location as the point A and the border of the pectoral muscle as a line BC. The skin air interface was identified as a curve passing through A and intersecting the pectoral muscle line. The nipple was defined as the origin of the coordinate system. A family of second order curves were defined through the nipple and intersecting the pectoral line (AD). Every pixel location in mammogram was represented by geodesic distance (s) from nipple and parametric angle (¿) as shown in figure 1. The scoring technique called MTR (mammographic texture resemblance marker) used this breast coordinate system to extract Gaussian derivative features. The features extracted using the (x,y) and the curve-linear coordinate system were used for MTR scoring in a case-control study (Otten et al, 2005) which includes mammograms (MLO view) of 245 patients diagnosed with breast cancer in the subsequent 2-4 years (123 interval and 122 screen detected cancers) and 250 matched controls. We evaluate the accuracy of MTR with and without the curve-linear coordinate system by computing the ROC curve. We assessed the use of the curve-linear breast coordinate system in registering the pairs of baseline and follow-up mammograms in a longitudinal study by comparing it with well-established registration measures such as mutual information, cross correlation and sum of squared difference in temporal study.

Result
Fig 1 shows the construction of anatomically oriented curve-linear breast coordinate system for a mammogram. MTR scoring with and without breast coordinate system is shown in table 1. It is worth noting that the proposed breast coordinate system was significantly better in risk segregation measured by the AUC compared to the traditional Cartesian x,y coordinate system (P<0.001, DeLong test). The curve-linear breast coordinate system provided the lowest registration dissimilarity measured compared to the other alignment methodologies as seen from table 2 in given temporal study.

Conclusion
The curve-linear anatomical breast coordinate system facilitated computerized analysis of mammograms. The proposed coordinate system slightly improved the risk segregation by Mammographic Texture Resemblance and minimized the geometrical alignment error. Potentially this coordinate system may be used also for improving precision in CAD systems.

Keywords: Breast Coordinate system, Breast Cancer, Imaging marker.

OriginalsprogEngelsk
Publikationsdato2011
StatusAccepteret/In press - 2011

Citationsformater