Feature extraction employing fuzzy-morphological decomposition for detection and classification of mass on mammograms.

Abstract

Mathematical Morphology is a theory of nonlinear processing extensively used in digital image processing. The computation effort, however, associated to Mathematical Morphology is usually large mainly because of high occurrence of conditional branches, due to use of intersection and union set operations. In terms of processing, it means the occurrence of maximum and minimum calculation. Mathematical Morphology, despite a relative large computational cost, is suitable to biomedical images applications, where both form and texture are essential in order to study anatomical deformities. We use detection and classification of mass in mammograms as a case study. The reason is that breast cancer is the leading cause of adult women by cancer worldwide. We propose a method inspired by series of wavelets for fuzzy-morphological decomposition in regions of interest on mammograms. Our decomposition employs nonlinear low-pass and high-pass filters based on openings and closings operations, which employ fuzzy-approximations. They replace conditional branches by arithmetic operators of subtraction and multiplication, computationally more efficient. We used 355 images of fatty breast tissue of IRMA database, with 233 normal instances, 66 benign, and 56 malignant cases. Classification was performed using SVM and ELM networks with modified kernels, in order to optimize accuracy rates, reaching 93.18%.