Evaluation of Segmentation Methods for Breast Cancer Detection

Abstract

Breast cancer is the most common cancer among women in the world. However, upon early detection, treatment can be carried out earlier and thus more efficient. A mammogram is the most common examination for early detection of this disease. There are various lesions characteristic of breast cancer such as tumors that can be detected on this mammogram. Computer Aided Detection (CAD) system operated in mammographic images is becoming increasingly popular in early detection of breast cancer. It’s mainly consists of pre-processing, segmentation, and classification steps. Segmentation is a very critical step in any mammographic CAD system which is considered a main step in defining the overall systems’ accuracy, so it’s a key driver that must be studied in depth in this research.There are many segmentation techniques in the literature that have addressed the problem of segmentation of mammograms. Thresholding, k-means clustering and level set algorithm were deeply addressed in this research. The main objective of this research is to systemically evaluate the three previously mentioned methods and determine which of the methods is recommended to be used for mammographic image segmentation. All the analysis was performed on fifty real case mammograms with different breast densities from mini MIAS Database and implemented using MATLAB. The methodology followed in this research consists of pre-processing, segmentation, post-processing an evaluation of the three methods. The pre-processing steps were the same in all three algorithms and consisted of adaptive histogram equalization followed by contrast enhancement. Then the image was fed to each of the segmentation algorithms individually namely, threshold, k-means clustering algorithm and level set algorithm, and then the post-processing steps were performed which consisted of area filtering and pectoral muscle removal from the segmented images. Finally, to evaluate the performance of the three methods accuracy, sensitivity and specificity measures were used. In conclusion it was found that the k-means clustering algorithm with 6 clusters was the most recommended technique for segmentation of breast tumours for mammographic images, the level set tends to produce better results in some images but it produces slightly lower results than the k-means, finally the binary threshold should be less likely used in this scope, hence it could be used when combined with a different segmentation method in order to produce better results.