Characterization of Mammogram Using Ensemble Classification Technique for Detection of Breast Cancer

Abstract

Breast cancer is one of the most common known cancers in women today. Just like any other form of cancer an early detection of cancer provides better chances of cure. However, it is an arduous task for the radiologists to detect cancer accurately. Thus computer aided diagnosis of the mammographic images is the most popular medium to aid the radiologists in accurately classifying benign and malignant mammographic lesions. In this thesis an efficient approach is presented to classify the mammographic lesion for the detection of breast cancer. In this approach the extracted feature coefficients are balanced using Gaussian distribution. This distribution balances the class unbalanced dataset providing for better classification. This scheme uses Logit Boost classification technique. Logit Boost uses least squared regression cost function on the additive model of Adaboost. The standard MIAS database was used to obtain the mammographic lesions. With a classification accuracy rate of 99.1% and a performance index value of AUC = 0.98 in receiver operating characteristic (ROC) curve the results are pretty much optimal. These results are very promising when compared with existing methods.