Breast masses diagnosis using supervised approaches

Abstract

Computer Aided Diagnosis (CAD) is usually used to assist radiologists while interpreting mammograms and help them improving breast cancer diagnosis accuracy at earlier stages. One of the main breast cancer early indicators is the presence of masses. CAD systems main target is to detect eventual masses from digital mammograms characterize them and evaluate their malignancy. In this paper, we introduce a new approach of breast masses diagnosis on digital mammograms, which begins with a preprocessing step where artifacts and pectoral muscle are removed and then the contrast is enhanced. The second step consists on segmenting breast masses, using Generalized Gaussian Density (GGD) estimation and a Bayesian backpropagation neural network. The last step is masses characterization using a combination of morphologic and textural features which are exploited to classify the segmented masses into benign and malignant classes, using a neuro-fuzzy system (ANFIS). The proposed CAD system was tested on the MIAS database and masses' detection rate has reached 97.08% with the GGD analysis and bayesian back-propagation neural network. 97% of these detected masses were correctly classified with an ANFIS system.