Development of breast lesion detection program

Abstract

Since the diversity of lesions represented in the breast and lesions that are shown in mammography were not clear in some cases and were similar to surrounding normal tissue which will cause to missing of diagnosis and specifying lesion localization, which resulted in difficulty of analysis in mammography. This study aimed to develop a program that use of Fuzzy C-means combined with morphological techniques to predict breast lesion. By using the sample of 213 mammographic images in Mediolateral Oblique view (MLO) with lesions type (Mass, Microcalcification or Normal) from the online database (The Mini-MIAS), the mammographs were divided into 3 groups (40 mammographs for Mass lesion, 17 Mammographs for Microcalcification lesion and 156 mammographic images for Normal which is not shown for lesion). The program was analyzed the pictures by using an algorithm that was set for specific lesion localization and predict the breast lesion represents in Mass, Microcalcification and Normal, and measured the sensitivity, specificity, and accuracy. The results show that sensitivity of mass and Microcalcification detection rate were 52.5% and 0%, respectively, specificity was 66.0% and accuracy was 63.2 and 59.5%, respectively.