Identification of Normal and Abnormal Mammographic Images Using Deep Neural Network

Abstract

This article presents an image retrieval strategy that considers the similarity of a selected image (CBIR-Content-Based Image Retrieval). The analogy is defined by the wavelet technique, combined with Hu moments, for removing features. The classification of mammography’s is conducted through the Artificial Neural Networks Auto-organizing System (SME). A Data Base (QUALIM), University of the Federal Republic of Sao Paulo (UNIFESP), and Medical Image Classification Laboratory are used to test the method. The system suggested is tested. We have employed two widely used pattern recognition approaches—facial recognition digital mammograms for examination. The techniques are based on the new classification schemes Ada Boost and Support Vector Machines (SVM).Several experiments were carried out to evaluate the accuracy of these two algorithms in different circumstances. The results for the Ada Boost classification system are positive, especially for mass lesions classification. In all cases, the algorithm was 76% exact, and only 90% accurate for mass. The SVM-based algorithm was not available. To improve the precision of the process, we need to choose enhanced image functionality for digital mammograms than those commonly used. Detection of breast cancer is the most challenging aspect in the field of health monitoring. This document has been used to evaluate breast cancer detection through a dataset of the mammographic image analysis company (MIAS). The suggested method has four key steps: preprocessing, segmentation, retrieval, and classification of images. Initially in mammograms, laplacian filtration was used to describe the edges area and was thus particularly susceptible to noise. The modified adjustable Fuzzy-C-MEANS (ARKFCM) was used in the following segmentation to find the object within the complicated module. The conventional ARKFCM masses of undefined mass were painful to divide into mammograms. The Euclidean distance in ARKFCM was replaced with the correlation function to solve this problem to increase its segmental efficiency. In the segmented cancer region, the removal of representative subsets was performed by extracting hybrid properties (histogram of guided grade (HOG), uniformity, and energy). Each feature value was defined for the Deep Neural classifier—network (DNN) for detection in normal and pathological areas of mammograms. The findings of the study show that the technique shows an improvement of up to 3-9% in precision compared with other methods currently in use in breast cancer classifications. Describes a new way to identify borders between different brightness areas. The goal is to create distinctions between regions that are unclearly distinct and defined by tonal shifts. The limits are set by the coordinates of the start and endpoints. The proposed method can be used as a stage in advanced techniques for hierarchical image analysis that increases the semantically awareness of picture content with each step. This study applies and evaluates mammograms. Interpretation of mammograms is a complex research area discussed by many authors. The treatment of breast cancer is achieved using these approaches.