Textural Feature Based Classification of Mammogram Images Using ANN

Breast cancer is the most common cancer in women worldwide which leads to death of the patient if not detected early. Breast cancer can be detected early with the help of mammogram images. Mammography is a standard imaging modality for the diagnosis and screening of breast cancer. However, suitable image processing techniques are required to detect such ambiguities. Textural analysis of such image is used to detect the cancer tissues. In this paper, we propose a method using Laws Texture Energy Measure (LTEM) as an approach for breast cancer detection. The LTEM method uses the energy maps of the feature matrix for the calculation of feature vector. A Back-propagation method using Artificial Neural Network (ANN) is used to classify the normal, benign and malignant tissue region. The mammography images are obtained from Mammographic Image Analysis Society (MIAS) database for experimentation and analysis. The proposed method is validated in comparison with different back-propagation algorithms and the results show the superiority of the proposed model.

I. Introduction

Breast cancer is the second most leading cause of death in women which can be avoided by effective detection and treatment at early stage. In medical practice, mammography is one of the suitable modality in the diagnosis process of breast cancer for recognizing the malignant tissues. However, it is difficult to recognize abnormalities in the breast tissues due to its low specificity nature. It is necessary to process and retrieve the information without affecting other feature which is present in the image. Thus an appropriate image processing technique is required for recognition of such abnormalities. There are number of approaches suggested by different researchers using texture analysis. Texture analysis is a coherent approach in image processing for breast cancer detection. Abnormality in mammogram images can be distinguished by analyzing their texture features without a detailed knowledge [5]. Several related research papers are discussed in the next section.