Multi-mass breast cancer classification based on hybrid descriptors and memetic meta-heuristic learning

Abstract

Among the causes of death in the world, breast cancer is considered the most common cause of mortality among women to the extent that one in five deaths among women is attributed to the incidence of this cancer. In this paper, we introduce a computer-aided detection approach to multiple classifications of breast masses. We tried to separate and intelligently recognize different masses in the breast cancer by means of mammograms so that in the first step, with the pre-processing, pectoral region is segmented from other parts and different areas are primarily clustered by K-means method. In the next step, using aggregation of efficient features such as texture features, Pseudo–Zernike moments, and wavelet features will be extracted from the input image and simulated annealing algorithm will reduce the size of feature vector. The final step will be the classification of possible masses in mammogram and the assessment of its severity based on memetic meta-heuristic adaptive neuro-based fuzzy inference system in which the optimizer is shuffled frog-leaping algorithm. The proposed method is evaluated using 322 mammogram images taken from Mini-MIAS database, which contain a variety of possible masses in mammograms. We compare our model with similar algorithms and several state-of-the-art methods through a comprehensive set of experiments. In this approach, the focus is on providing a hybrid algorithm for accurate detection and extraction of masses in mammography, with the approach that the physician can predict both the potential disease stage and type of tumor.