Certain Investigations On Mammographic Abnormalities Using Hybrid Feature Based Machine Learning Techniques

Abstract

Cancer is a group of diseases that causes cells within the body to newlinechange and grow out of control Breast Cancer BC is one of the leading newlinecauses of death among women who ails from cancer The early detection and newlineaccurate diagnosis of breast cancer is still an unresolved challenge in the newlinemodern computer aided detection analysis Even though biopsies are taken newlinetumors frequently go undetected until a stage where therapy is costly or newlineunsuccessful The exact detection of suspicious breast cancer region in newlinewomen is ambiguous in many cases newlineThe major reasons for failure of automatic detection in mammogram newlineare due to segmentation and classification of suspicious region It is still a newlinechallenging task to segment the abnormalities due to distribution with newlinevarying intensity noise in image acquisition and ambiguity in anatomical newlinestructures of mammograms The three main lesion features are texture feature newlineshape feature and gray level feature in mammogram The detection of newlinemammographic abnormalities found in literature of last decades is mainly newlinecharacterized by single or double features Algorithms used in these newlineliteratures obtain good detection results on one type of lesions but it may newlinegenerate unreasonable detection results on other types of lesions To newlineovercome these limitations new algorithms for the detection of masses and newlinemicrocalcifications have been proposed in this research study The proposed newlinealgorithm consists of the combination of bootstrap probabilistic techniques newlinewith classical segmentation approach hybrid feature extraction and support newlinevector machine classifier for detection of masses and microcalcifications newlineThe main aim of this thesis is to improve the detection accuracy of newlinebreast abnormality in mammogram A new framework for the detection of newlinelesions of any shape margin and size using a single mammographic view is newlineproposed With the aim of detecting the masses two methods are proposed newlineviz KMeans Bootstrap Subgroup KMBS and Expectation Maximization newlineBootstrap Subgroup EMBS With the aim of detecting the cluster of newlinemicroca