An Automatic Statistical Method to detect the Breast Border in a Mammogram

Abstract

Segmentation is an image processing technique to divide an image into several meaningful objects. Edge enhancement and border detection are important components of image segmentation. A mammogram is a soft x-ray of a woman's breast, which is read by radiologists to detect breast cancer. Recently, digital mammography is also available. In order to do computer aided detection on mammogram, the image has to be either in digital form or digitized. A preprocessing step to a digital/digitized mammogram is to detect the breast border so as to minimize the area to search for breast lesion. An enclosed curve is used to define the breast area. In this paper we propose a modified measure of class separability and used it to select the best segmentation result objectively, which leads to an improved border detection method. This new method is then used to analyze a test set of 35 mammograms. The breast border of these 35 mammograms was also traced manually twice to test for their repeatability using Hung's method1. The borders obtained from the proposed automatic border detection method are shown to be of better quality than the corresponding ones traced manually.

Keywords

• Mammogram
• Breast border
• Area tracing
• Validation
• Separability