Single Layer Neural Network Based Model for Mammogram Mass Segmentation and Classification

Abstract

newline Breast cancer early detection and diagnosis depend heavily on the segmentation and classification of mammograms. This study explores the application of the Functional Link Artificial Neural Network (FLANN) to enhance these processes. Mammograms present significant challenges for traditional segmentation and classification techniques due to their excessive noise and unpredictability. Each year, a large number of new studies describing novel algorithms for the classification and segmentation of mammogram masses are published in the literature. Nevertheless, the variability of tumor size, shape, texture, appearance, and contrast poses significant obstacles for the detection of masses through mammography. newlineThis thesis makes an attempt to address many issues related to the segmentation and classification of mammography masses. As a consequence, a novel framework for suspicious lesion pre-processing, segmentation, and classification is proposed. Firstly, several denoising methods for mammogram images are compared. Second, a new thresholding-based method that uses FLANN is suggested for identifying questionable lesions in mammogram images. It automatically updates threshold value determined by the attributes of the image. To achieve this, FLANN using Block Based Normalized Sign Sign Least Mean Square as the algorithm for training is used, to obtain best threshold values. Thirdly, an innovative segmentation technique utilizing a hybrid model with a Particle Swarm Optimization (PSO)-optimized FLANN is used. Adaptive threshold change depending on image attributes is a significant feature of the segmentation technique, suggesting a sophisticated and adaptable method for mass segmentation in mammography. Finally, two approaches are discussed for mammogram mass classification. First, a two- stage FLANN model is suggested for segmentation and classification masses in mammograms. By changing the input space and permitting more accurate border delineation, the first-stage FLANN is used to segment the masses.