Classification of breast lesions using dynamic and morphological features in DCE-MRI

We investigate the performance of several classifiers using dynamic and morphological features for breast lesions classification in Dynamic Contrast Enhanced-Magnetic Resonance Imaging (DCE-MRI). 22 malignant and 20 benign breast lesions, histologically proven, were analyzed. Four different classifiers were used: a Multilayer Perceptron, a Support Vector machine, a Bayes classifier and a Decision Tree classifier. 98 dynamic and 54 morphological features were extracted from the Volumes of Interest (VOIs) that have been manually segmented by an expert radiologist on dynamic images of the lesions. The performance of the classifiers has been compared with the histological classification of lesions. Results indicate that the Bayes classifier presents the better results in term of sensibility, specificity and accuracy when morphological features were utilized and the Decision Tree when dynamic features were utilized. The best results were obtained using morphological feature in comparison to dynamic feature