Analysis of Different Fuzzy C-Means Membership Functions for the Design of CAD Tool of Early Breast Cancer Detection

Abstract

In this paper, analysis of the different membership function for fuzzy C-means based segmentation technique are done to overcome the issue of random initialization of membership matrix, 116 hybrid features set and analysis of various classifier with modified K nearest neighbor (KNN) are proposed for primary breast cancer detection. The steps involved to design CAD tool for breast cancer detection includes preprocessing, segmentation, feature extraction, selection, and classification. In pre-processing, manual cropping of original mammograms for removal of background details, and contrast enhancements are used. Contrast Limited Adaptive Histogram Equalization (CLAHE) is used for enhancement of mammographic images. Further, nonlinear complex diffusion based unsharp-masking and crispening method is utilized to highlight the abnormalities of mammogram such as micro calcifications, tumors, etc. Further, 2D DWT is applied for taking approximation coefficient followed by fuzzy C-means segmentation with different types of fuzzy membership functions, which is utilized to segment and extract the region of interest. For feature extraction, 116 hybrid features are extracted from each image such as Geometric based features, Texture based features, Gabor based features and Histogram based features. Next step is Minimum Redundancy and Maximum Relevancy (MRMR) is used as features selection technique for discarding repetitive and irrelevant features. Final step is classification, the modify KNN classifier based on mutual information in place of Euclidean distance is proposed while finding the nearest neighbors. The comparative analysis of the fuzzy C-means segmentation techniques with different types of fuzzy membership methods are presented in terms of RI, GCE, VOI and the various classifier performance are taken in terms of accuracy, precision, sensitivity, specificity.