Early detection of Breast Cancer using Computational Intelligence Methods

Abstract

The thesis presents computer-aided diagnosis methods to assist radiologists in the early, reliable, economic, fast, and reasonable diagnosis of breast cancer utilizing mammograms. As there is a large disparity in the pixel intensities or grey levels in a mammogram, representation of certainty/uncertainty is adopted using the concept of information set while developing methods for the feature extraction, classification along with enhancement of different categories of breast cancer, and learning of the unknown parameters involved in the methods. newline newlineThe information set enlarges the extent of a fuzzy set in the sense that a compliment membership function has a part to play. Each component in a fuzzy set is a pair consisting of an attribute value named as information source value and its resultant membership function value whereas the information set consists of information values as its elements. The Hanman Anirban entropy function relates the constituents of each pair with a product named as the information value while delineating the certainty/uncertainty in a fuzzy set. The sum of information values in a fuzzy set gives the extent of possibilistic certainty associated with a concept or class. An extension of the information set is carried out in conformity with an intuitionistic fuzzy set to create the concept of pervasive information set in which the pervasive membership function is an amalgamation of a non-membership function and membership function both of which address the inaccurate fuzzy modeling. newline newlineThree feature extraction methods are developed involving firstly the basic information set-based features, secondly, the pervasive information set-based features, and thirdlypervasive texture information set features. Two classifiers called the Hanman transform classifier and hesitancy-based Hanman transform classifier are formulated for accomplishing multi-class classifications. New deep learning systems called HanmanNets allowing the modification of Kernel functions as well as feature maps of ResNet architectures