Multiresolution and Multiscale Geometric Analysis based Breast Cancer Diagnosis using weighted SVM

This paper presents an approach for breast cancer diagnosis in digital mammogram using multiresolution and multiscale geometric analysis. The proposed method consists of two stages. In the first stage, mammogram images are decomposed into different resolution levels using wavelet transform and curvelet transform, which are sensitive to different frequency bands. A set of the biggest coefficients from each decomposition level is extracted as features vector. In the second stage, classification is performed on a weighted support vector machine (SVM). Due to random selection of samples, it is highly probable that a significantly small portion of the training set is the "mass present" class. To address this problem, we propose to use weighted SVM in a successive enhancement learning scheme to examine all the available "mass present" samples. The proposed approach is applied to the Mammograms Image Analysis Society dataset (MIAS) and classification accuracy of 99.3% is determined over an efficient computation time by successive learning enhancement. Experiment results illustrate that the multiresolution and multiscale geometric analysis-based feature extraction in conjunction with the state-of-art classifier construct a powerful, efficient and practical approach for breast cancer diagnosis.