Evaluating the Effectiveness of Classical Denoising Filters for Microcalcification Segmentation in Mammogram

Abstract

One of the vital signs of early breast cancer is calcification in breast masses. Although calcification has been a principal indicator of malignant tumor in breast imaging treatment, accurate detection and interpretation of calcification using mammograms remain challenging, which is mainly due to their tiny sizes, heterogeneous structures and background noises. Often the secretive nature of calcifications reduces the credibility of efficient diagnosis, and in most times it is difficult to identify whether calcifications are benign or malignant. Therefore, extracting the breast region precisely from a mammogram is an essential step for automatic segmentation during computer-aided diagnosis and classification of disease state. In this study, we have evaluated the efficacy of three most widely used image denoising methods (median filter, Gaussian filter and Weiner filter) for fast and accurate detection of micro- or macro-calcification using Otsu-thresholding technique, using mammogram images of MIAS database. This systematic analysis revealed a simple and accurate image processing framework for the detection of heterogeneous calcification in Medio Lateral oblique (MLO) view of breast mammograms. For performance evaluation, three different classical filters were applied to various mammogram images with heterogeneous breast densities (fatty, dense, dense granular). The efficacy of the present method was verified based on several performance indices and found to be similar to that of other complex segmentation techniques. Overall, these results show a simple and efficient way of segmenting mammogram calcification using real data and can be applied to analyze other abnormalities in mammogram images for breast cancer diagnosis.