Enhancing the Mammogram image Classification using Mutual Information Feature Selection

Abstract

The Breast Cancer is one of the main causes of death for women all over the world. With early and accurate diagnosis of the breast cancer the cure rate rises from 56% to more than 86%. The objective of this study is to enhance the classification accuracy of the mammograms images based on feature selection method to detect if the input image is normal or affected by the diseases. The accuracy of most of the classification methods depend on important features extracted from the mammogram images and the classifier itself. This study propose a classification method based on K-Nearest Neighbor (KNN) and Support vector machine (SVM) using important features selected from data set of features extracted from the mammogram images The MIAS Mini data set (Mammographic Image Analysis Society) include 209 normal images, 23 images of CIRC (Circumscribed masses), 19 images of SPIC (Speculated masses),19original images of MISC(ill-defined masses), 23 images of CALC(Calcification)) based on mutual information (MI) feature selection method.in this study the classification process includes five basic steps; beginning with the mammogram Image collection, image processing, features extraction , classification and testing and evaluation ;firstly by using all features, secondly by using more important features based on mutual information (MI) features selection method, the last step is testing and evaluation. This study used set of thirteen features, extracted from mammogram images that taken from MAIS database, then it applies K-nearest neighbors (KNN) and Support vector machine (SVM) based classification method. In this study, the dataset splited into two parts, namely: training and testing. After the construction of the classifier based on training data, the proposed model using the test data to measure the accuracy. The best accuracy obtained was 83% by KNN algorithm when using percentage of 85% and 15% for training and testing by using the most important features for the five Sub-Features best features. Using other feature selection method may results in more accuracy of the classifier