Copyright © 2014 American Scientific Publishers All rights reserved Printed in the United States of America RESEARCH ARTICLE Journal of Medical Imaging and Health Informatics Vol. 4, 1–7, 2014 Classification of Breast Tissue as Normal or Abnormal Based on Texture Analysis of Digital Mammogram Fatehia B. Garma and Mawia A. Hassan ∗ Biomedical Engineering Department, Sudan University of Science and Technology, Khartoum, Sudan The breast cancer is a serious public health problem among women in the world. Efforts in Computer Vision have been made in order to improve the diagnostic accuracy by radiologists. In this paper a method for detection of breast cancer based on digital mammogram analysis was presented. Haralick texture features were derived from spatial grey level dependency (SGLD) matrix. The features were extracted from each Region of interest (ROIs). The features discriminating to detect abnormal from normal tissues were determined by stepwise linear discriminant analysis classifier. The proposed method achieved 95.7% of relative accuracy for classification of breast tissues based on digital mammogram texture analysis. Keywords: Breast Cancer, Mammogram Image, Texture Analysis, Haralick Texture Features, Linear Discrimination Analysis. 1. INTRODUCTION Breast cancer is a type of cancer originating from breast tissue. It is the most public health problem among women. If breast can- cer is detected early; the treatment can be performed earlier and therefore be more efficient. Mammography is the most common technique for early detection of breast cancer. It is considered the most effective, low cost, and reliable technique for early detec- tion of this disease. 1 With the advances of digital image processing, radiologists have a chance to improve their performance with computer- aided detection and diagnosis (CAD) system; this technology can be used with standard film mammograms or with digital mammograms. 2 Breast Cancer cells can be of different types and shapes. 19 The presence of other structures makes the mammogram background very complex for physician to distinguish malignant mass lesions from normal breast tissues. 12 Also, the sensitivity of mammo- graphic screening varies with image quality and expertise of the radiologist. 20 That leads to misinterpretation of mammograms, and to reduce the high misinterpretation rate, an objective method to classify and identify the pathology on the mammogram is needed. 21 One method to identify the breast cancer is texture analysis in mammogram. Textures are one of the important characteris- tics for identifying objects and ROI of various images. 3 Texture ∗ Author to whom correspondence should be addressed. analysis is important for application of computer image analysis for classification, detection and segmentation of an image based on intensity and colour. 4 2. PREVIOUS METHODS Several papers addressed the issues involved to detect and clas- sify the breast cancer in digital mammograms. Chan et al. 2 clas- sified breast tissue on mammograms into masses and normal using texture features derived from SGLD matrix and stepwise linear discriminant analysis to perform classification. The clas- sifier achieved an average area (Az) under the receiver operat- ing characteristics (ROC) curve, Az = 084 during training and 0.82 during testing. Wang and Karayiannis 5 were presented an approach for detecting microcalcifications in digital mammo- grams using wavelet-based sub-band image decomposition, and then reconstructing the mammogram from the sub-bands con- taining only high frequencies. The reconstructed mammogram is expected to contain only high-frequency components, includ- ing the microcalcifications. Sheshadri and Kandaswamy 6 stud- ied breast tissue classification using statistic feature extraction of mammography. The statistical features extracted are the mean, standard deviation, smoothness, third moment, uniformity and entropy which signify the important texture features of breast tissue. Classify the breast tissue into four basic categories like fatty, uncompressed fatty, dense and high density. The Accu- racy of the proposed method has been verified with the ground truth given in the data base (mini-MIAS database) and has J. Med. Imaging Health Inf. Vol. 4, No. 5, 2014 2156-7018/2014/4/001/007 doi:10.1166/jmihi.2014.1310 1