Accurate skin-line estimation is an important prerequisite for enhancement and analysis of mammograms for computer-aided diagnosis (CAD) of breast cancer. Appropriate display of benign and malignant lesions inside the breast region in mammograms has a direct relationship to the skin line of the breast. Even a small windowâlevel operation on a digital display by a radiologist to enhance the visibility of lesions inside the breast region can lead to diminished skin-line information due to the loss of gray-level contrast. In this process, a lesion located near the skin-line may also be lost. This is attributed to the fact that the breast tissue in the skin-line zone is less dense compared to the other tissues in its neighborhood (see Fig. 11.1). Although the above fact poses an inherent challenge, there are other challenges such as noise present in low-contrast mammograms. The noise can come from digital- or film-acquistion processes called "system noise." Besides the structural and systemic challenges, automatic skin-line estimation is hindered by extraneous structures in x-ray mammograms such as labels for patient identification, especially when the labels are placed close to the breast. It is proposed to solve such difficult cases by our novel automatic skin-line detection strategy. The variability present between the images in a data set due to breast size and protocols of image acquisition can make a skin-line detection system complex. Having discussed the prominent challenges posed in automatic breast skin-line estimation, previous related works by other research groups are reviewed.
After carefully evaluating the strategies used by other research groups for skin-line estimation, the proposed algorithms are classified into two categories: (1) histogram-based thresholding, and (2) the boundary-based deformable model. The technique is novel: a combination of knowledge regarding the anatomy of the breast, and constraints between the skin-line boundary and the edge of the stroma of the breast are introduced. In essence, the algorithm is a hybrid approach with several novelties.
Online access to SPIE eBooks is limited to subscribing institutions.