Classification using fractal features of well-defined mammographic masses using power spectral analysis and differential box counting approaches

Abstract

Computer-aided diagnosis (CAD) of mammograms assists the radiologists to detect the mammographic mass presence. Since the variability of shapes of such breast masses occurs very frequently, it is more difficult to classify them into benign and malignant stages. One efficient approach to classify them is the fractal analysis by deriving their shape features. Various methods have been proposed for the fractal dimension (FD) computation of region of interest (ROI) in biomedical images. Among those, two methods, namely power spectral analysis (PSA) and differential box counting method (DBCM), are used here for the FD computation of breast contour margins. Fractal analysis by PSA method is a frequency-domain approach which is applied to the one-dimensional (1D) signatures of the two-dimensional (2D) breast mass contours. However, the DBCM model assigns the smallest number of boxes that cover the whole image surface. Finally, a comparative analysis is performed between the above-said two methods which show the PSA method yields better accuracy than the DBCM.