Detection of Microcalcification clusters in Breast mammograms by using Neuro-fuzzy technique Sareh Habibi 1 and Abbas Karimi 1* 1 Department of Computer Engineering, Faculty of Engineering, Arak Branch, Islamic Azad University, ARAK, Markazi, IRAN * arak.ac.ir - akaimi@iau Abstract Breast cancer is one of the most common cancers among ladies entire the world and its early stage detection has a considerable effect on the survival rate of patients. Breast cancer is exposed by microcalcification clusters in form of tiny specks in bright colors in early stages of disease. In this study, a diagnosis model based on Neuro-fuzzy technique is presented This method used local statistics for computing threshold value rather than previous method that need manual setting of threshold to detect microcalcifications in mammograms. It classifies normal mammograms from mammograms containing malignant microcalcification clusters. The results showed that rates of true-positive and true-negative of the proposed method reached up to 96.64% and 36.3% respectively. Moreover, sensitivity of 95% achieved that is quite high in comparison with maximum 84% for an experienced radiologist. Keywords: Breast cancer, microcalcification clusters, mammogram images, Neuro-fuzzy technique. 1. Introduction Breast cancer is the most important cancer type among the women all over the world and comprises 25% of all cancer cases. The rate of survival of this disease depends on type, disease progress and patient's age [1]. The survival rate of patients in developed counties including US and UK in 5 years period is 80% and 90% respectively, while it is less in developing countries[2], [3]. Breast cancer is verified by biopsy of tumors. Fortunately a considerable rate of breast cancer cases among ladies is detectable by self-examination during menstrual period. However, advanced cases of breast cancer are normally detected late due to patients neglect, cultural limitations or un-sufficient educations. Mammography is the most common and economic way for screening the breast tissue in order to detect breast tumors or lumps. This approach was reported to increase diagnosis of breast cancer by 16% [4], moreover, CAD systems based on mammogram improved detection sensitivity of a junior radiologist from 61.9% to 84.6% [5]. Reading of mammograms by a CAD system and an experienced radiologist were compared in [6] showing that comparable results were obtained from the CAD system with the experience radiologist. Microcalcifications are small calcium deposits that look like white specks on a mammogram. Microcalcifications are usually not a result of cancer; however breast cancer is exposed by microcalcification clusters by specific patterns in bright colors in early stages of disease. These clusters are categorized by using a lexicon of BI-RADS (Breast Imaging Reporting And Data System) by which medical team can reach a conclusion on the presence or stages of breast cancer [7]. Based on the lexicons, microcalcifications are definitely counted as benign and there is no need to biopsy when they are eggshell or rim-like, coarse and popcorn-like, vascular, rod-like, teacup-shaped, or lucent-centered. They are in medium concerns when calcifications are tiny and hazy. However, linear, fine, branching or casting structure, or no certain shape and structure of calcifications are signs of malignant tumor[7]. Mammogram pleomorphic calcifications and fine, linear, branching calcifications as signs of malignant calcifications are delineated in Figures 1(top) and 1(bottom), respectively. Gray level of the image represents the density of a given area of breast while the contrast may show a potential tumor. In this case, median filter, variance and contrast of the area are taken into account. Selecting proper features for extraction in the mammograms and applied approach for clustering have a significant effect on the quality of the image and the presentation of the tumors [8]. International Journal of Computer Science and Information Security (IJCSIS), Vol. 15, No. 1, January 2017 651 https://sites.google.com/site/ijcsis/ ISSN 1947-5500