Artificial Intelligence Review | 2019
A review of breast boundary and pectoral muscle segmentation methods in computer-aided detection/diagnosis of breast mammography
Mammography can be considered as the current gold standard for detecting early signs of breast cancer and is in wide use throughout the world. As confirmed by many studies, breast cancer screening using mammography can reduce breast cancer-related mortality by 30–70%. However, although the interpretation of mammography images by a second reader has been shown to increase the cancer detection rate, this practice is not widespread due to the cost associated. As a result, computer-aided detection/diagnosis (CAD) of breast mammography has been gaining popularity with various studies illustrating the positive effects of using computers in detecting early breast cancer signs by providing the radiologists with a second opinion with most of these CAD systems requiring the breast outline and pectoral muscle regions (in images acquired using Medio-Lateral-Oblique view) to be segmented from mammograms prior to the classification. This paper discusses recent developments and methods proposed for segmenting the breast and pectoral muscle regions and compares the performance and shortcomings of different approaches grouped together based on the techniques used. While it is arduous to compare these methods using comparative analysis, a set of common performance evaluation criterion is defined in this study and various methods are compared based on their methodology and the validation dataset used. Although many methods can achieve promising results, there is still room for further development, especially in pre-processing and image enhancement steps where most methods do not take the necessary steps for ensuring a smooth segmentation of boundaries. In this paper, the most effective pre-processing, image enhancement and segmentation concepts proposed for breast boundary and pectoral muscle segmentation are identified and discussed in hopes of aiding the readers with identifying the best possible solutions for these segmentation problems.