Various registration approaches have been adopted to correlate areas in mammograms and detect differences. These differences are likely to be associated with the development of an abnormality. Most of these methods rely on corresponding landmarks between images. Obtaining these landmarks turns out to be the most difficult task and plays an important role in the alignment accuracy. Manual landmark generation is a tedious and time-consuming task when the number of landmarks is large and, moreover, introduces variability in placing the landmarks (two consecutive experiments with the same images could have different landmarks and errors). Therefore, a large number of these methods are based on automatically extracting and corresponding landmarks from mammographic images. The landmarks include breast boundary, pectoral muscle, salient regions extracted using wavelets, isointensity contours or steerable filters, and crossing points of horizontal and vertical structures.
This chapter presents an approach to extract correspondence in mammographic images based on anatomical features that appear as linear structures (i.e., fibrous tissue, ducts, and vessels). Correspondence is used here in two different applications: for mammographic image registration and for tracking linear structures.
Registration is applied to temporal and contralateral mammograms of the same woman. A framework for the detection of abnormal structures based on registration is developed, and some initial promising results are given. Tracking of linear structures (see Section 14.6) is performed for the same breast over several screening rounds. This could be used to assess and model the development of architectural changes and abnormal structures. By being able to track regions back in time, the available information helps to improve early detection of subtle abnormalities that are initially missed by radiologists. One could argue that contralateral registration based on the linear structures in the left and right breasts should not be performed because linear structures present large differences. However, it is generally accepted that overall duct, vessel, and tissue structure shows similarities between left and right breasts of the same woman. Figure 14.1 illustrates those similarities in temporal and contralateral images. The aim is to use these similarities to guide the registration process.
Online access to SPIE eBooks is limited to subscribing institutions.