Breast abnormality detection in mammograms using Artificial Neural Network

Breast cancer is one of prevalent diseases in women, and can be diagnosed using several tests that include mammogram, ultrasound, MRI and biopsy. Over the years, the use of learning machine and artificial intelligence techniques has transformed the process of diagnosing breast cancer. However, the accurate classification of breast cancer is still a medical challenge faced by researchers. Difficulties are routinely encountered in the search for sets of features that provide adequate distinctiveness required for classifying breast tissues into groups of normal and abnormal. Therefore, the aim of this study is to propose a system for diagnosis, prognosis and prediction of breast abnormality using Artificial Neural Network (ANN) models based on two dimensional wavelet transform. Three major approaches were taken in this study; preprocessing, wavelet decomposition analysis and neural network approaches. The first approach entails the preprocessing step for breast profile extraction, carried out by eliminating the low frequency components of the mammogram, leaving behind subbands containing high frequency coefficients, based on the idea that microcalcifications signify high frequency coefficients. The next approach involves features extraction derived from wavelet decomposition analysis. The final approach is referred to as the classification stage that utilizes back propagation neural network to distinguish abnormal tissue from normal ones. The proposed system was tested on the MIAS database, resulting in 91.64% succession rate of classification.