Analysis of texture for classification of breast cancer on mammogram images

Abstract

Breast cancer is a top cancer among women in the world. In conventional method, breast cancer can be detected by a medical expertise observation on patient's mammogram images. However, this method could lead to misdiagnose in distinguishing an interest object with naked eyes due to low quality of images. This research aims to classify mammogram images into three classes, i.e. normal, benign and malignant based on texture features. Some pre-processing techniques were involved, including removing the artefacts, cropping breast area, contrast enhancement and smoothing with median filter. Afterwards, some texture features were extracted followed by classification process by using multi-layer perceptron (MLP) classifier. Classification of normal and abnormal successfully achieved an accuracy of 98.33%, sensitivity of 100% and specificity of 97.5%. Whereas, for classification of three classes (normal, benign and malignant) achieved an accuracy of 90%, sensitivity of 85% and specificity of 87.5%.