An automatic mass detection in digital mammogram

Abstract

Breast cancer is one of the leading causes of the death of middle-aged women all over the world. Therefore, the detection of malignant masses is an important approach to reduce the mortality rate caused by breast cancer. Although several methods for mass detection have been studied, it is hard to obtain a good detection because of some characteristics of mass such as obscured shape and variation in the intensities of the masses. In this thesis, I propose a new method to detect malignant masses using adaptive segmentation method based on a combination of binary decision tree threshold method and global threshold method. By image analysis, I also propose two features based on measuring centroid distribution: normalized number of centroids and normalized variance of centroid distribution. Finally, I use binary decision tree classifier to classify whether input mammogram is benign or malignant. By using Mammographic Image Analysis Society (MIAS) database in experiments, my simulation results show that the accuracy of all system of my proposed method is 93.1%. Furthermore, the True Positive detection rate of my proposed method is 85.71% for a data set of 58 mammograms with tumors in each of them.