Efficacy of feature selection techniques for Multilayer Perceptron Neural Network to classify mammogram

Abstract

Mammography aims to detect, characterize and evaluate the findings of breast cancer and related breast diseases. An X-ray of a patient's breast is a diagnostic mammogram which can show anomalies being detected through screening mammography or earlier mammography findings requiring follow-up. In automated mammogram classification, features are extracted and classified. The number of features is high, and the feature set contains irrelevant and redundant features leading to degradation of classifiers. In this work, the features from the mammograms are extracted using Multi-scale filter bank and Genetic Algorithm (GA) based feature selection is evaluated. This study proposes to classify features using Multilayer Perceptron Neural Network (MLPNN). Results show improvement of the proposed method.