220 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) ISSN (Print) 2313-4410, ISSN (Online) 2313-4402 © Global Society of Scientific Research and Researchers http://asrjetsjournal.org/ Preprocessing of Digital Mammogram Image Based on Otsu’s Threshold Ashgan M. Omera*, Mohammed Elfadilb aBiomedical Engineering Department, College of Engineering Science, Sudan University (SUST),74 Khartoum, Sudan bCollege of Medical Radiological Science, Sudan University (SUST), 407 Khartoum, Sudan aEmail: ashganomer@yahoo.com bEmail: mohamedelfadil@gmail.com Abstract Mammograms are difficult images to interpret. Hence, a preprocessing stage is very important to standardize the mammogram image along with the reduction of its size, and improve the quality of image in order to produce reliable image for CAD system. The proposed technique of preprocessing involves removal of unwanted parts from background of the mammogram, removal of pectoral muscle, and image enhancement. Binarization based on Otsu’s threshold is a main process in all preprocessing steps. Multi-level thresholding applied to segment the pectoral muscle, and level three shows perfect results of pectoral muscle segmentation. A propose method applied on 160 images from MIAS database. Using of level-three multi-thresholding technique, the success rate was 96% in mammogram preprocessing stage. Keywords: Mammogram; Binarization; Otsu’s Threshold; Multi-level Threshold. 1. Introduction Preprocessing is very important stage aid to limit the search for abnormalities on only breast region without influence from background. Preprocessing helps in reduction of mammogram size, and improves the quality of the images to make the feature extraction phase more reliable. ------------------------------------------------------------------------ * Corresponding author. http://asrjetsjournal.org/ American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2017) Volume 37, No 1, pp 220-229 221 A mammography images involves excessive parts, these parts are not important in processing of mammogram image. Some parts of images have some notes and labels which consist of information like name, date, etc., these labels are bright as gray level of some breast tissues which will influence the classification and will not give the desired results [1]. The presence of noises and artifacts can disturb the detection of breast cancer and reduce the rate of accuracy in analysis [2]. Removing the excessive parts, all irrelevant information, noises and artifacts are the aim of preprocessing. As the mammogram images are texture in nature, it is very important to get proper enhancement for these images. In low contrast images, the minor difference between normal and the malignant tissue is not discernable and make the interpretation very difficult [3]. Hence, preprocessing enlarge the intensity difference between objects and background to produce reliable image for processing step. In mammograms, the pectoral muscle has nearly homogeneous gray level values and exhibits as a high intensity region. Failure in segmenting the pectoral muscle may cause a higher number of false positives for breast cancer detection [4]. So a method represented to automatic pectoral muscle removal from mammogram images. Numerous papers introduced different approaches for preprocessing mammogram images. Researchers used various methods to segment breast region and eliminate the background and pectoral muscle. In [5] are introduced a preprocessing technique based on threshold to omitting excessive sides and put all images in one side, then eliminated labels and background based on region growing method. The result obtained 99% by applied on 60 images of MIAS database. In [6], pectoral muscle removed by using watershed transformation and merging algorithm. The method applied on 84 mammograms from MIAS database, the results obtained are 0.85% and 4.88% for mean false positive and mean false negative rates respectively comparing with manually identified pectoral muscle. Researchers in [7] applied three steps on MIAS database. First step is contrast enhancement then pectoral muscle detection and suppression, they used CLAHE technique and seeds region growing. Out of 322 mammogram images, fourteen images failed in output; the accuracy obtained was 95%. In 2016, a study applied on all MIAS database for pectoral muscle segmentation. Pectoral muscle detected by using the morphological method and the random sample consensus (RANSAC) algorithm. The results showed 92.2% accuracy [8]. 2. Materials and Methods To achieve the preprocessing, the proposed method consists of three stages, unwanted regions removal, pectoral muscle removal, and image enhancement. However, the binarization consider as a main process in all preprocessing stages. 2.1. Image Database The mammogram images used for this study are from the Mammographic Image Analysis Society (MIAS) [9]. The MIAS database includes 320 films only mediolateral oblique (MLO) views taken from the UK National Breast Screening Program and digitized originally to 50µm/pixel and a gray-scale resolution to 8 bits per pixel (bpp), with a Joyce-Loebl scanning microdensitometer SCANDIG-3. The original mammograms are 1024×1024 American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2017) Volume 37, No 1, pp 220-229 222 pixels. The mammogram images are investigated and labeled by expert radiologists. 2.2. Image Binarization Binarization is a process, which converts the grayscale image to binary scale image. A binary images (I) is an image in which the pixels are only allowed two possible intensity values, nominally zero and one as in equation (1). Binarization of the image is very important for separating the foreground object from the background. The binarization methods can be categorized in different groups according to the threshold calculation method [10]. 𝐼𝐼 ∶ (𝑥𝑥,𝑦𝑦) → {0,1} (1) 2.2.1. Thresholding Thresholding is used to extract an object from its background by assigning an intensity value for each pixel, so that pixel is either classified as an object or a background [11]. Global thresholding use an appropriate threshold (T), as in (2). 𝑔𝑔(𝑥𝑥,𝑦𝑦) = �1, 𝑖𝑖𝑖𝑖 𝑖𝑖(𝑥𝑥,𝑦𝑦) > 𝑇𝑇 0, 𝑖𝑖𝑖𝑖 𝑖𝑖(𝑥𝑥,𝑦𝑦) ≤ 𝑇𝑇 (2) Threshold-based algorithms are divided into single-level and multi-level thresholding categories. 2.2.1.1. Otsu’s thresholding Otsu’s method [12] is the most successful global thresholding method. It automatically performs histogram shape based image thresholding for the reduction of a gray level image to a binary image. The algorithm based on image variance, Otsu’s method chooses optimal threshold by maximizing the between class variance (3). The optimal threshold value (t*), satisfies: 𝜎𝜎𝐵𝐵2(𝑡𝑡∗) = 𝑚𝑚𝑚𝑚𝑥𝑥0<𝑡𝑡<𝐿𝐿−1 𝜎𝜎𝐵𝐵2(t) (3) Where: 𝜎𝜎𝐵𝐵 between class variances, L intensity levels[0, … , 𝐿𝐿 − 1]. 2.2.1.2. Multiple thresholding The multilevel threshold segments the pixels into several distinct groups in which the pixels of the same group have gray levels within a specific range (4). The Otsu’s method can be applied for multiple thresholding segmentation [13], [14], the optimal thresholds 𝑡𝑡1∗ and 𝑡𝑡2∗ can be computed as equation (5). 𝑔𝑔(𝑥𝑥,𝑦𝑦) = � 𝑚𝑚, 𝑖𝑖𝑖𝑖 𝑖𝑖(𝑥𝑥,𝑦𝑦) > 𝑇𝑇2 𝑏𝑏, 𝑖𝑖𝑖𝑖 𝑇𝑇1 < 𝑖𝑖(𝑥𝑥,𝑦𝑦) 𝑐𝑐, 𝑖𝑖𝑖𝑖 𝑖𝑖(𝑥𝑥,𝑦𝑦) ≤ 𝑇𝑇1 ≤ 𝑇𝑇2 (4) American Scientific Research Journal for Engineering, Technology, and Sciences (ASRJETS) (2017) Volume 37, No 1, pp 220-229 223 𝜎𝜎𝐵𝐵2(𝑡𝑡1∗, 𝑡𝑡2∗) = max 0<𝑡𝑡1<𝑡𝑡2