Methods of selecting contours in multi-scale analysis of medical images.

Abstract

A brief overview of existing image segmentation methods based on contour selection is given. The forms of operators that are widely used in medical image processing are considered, the characteristics of these methods, their advantages and disadvantages are given. In parallel with these methods, wavelet analysis of images is considered for compression of their total volume. As an example, Maal's algorithm is chosen, which is based on the application of quadrature mirror filters in the decomposition and restoration of images at each stage of processing. The MSEC (Multi Scale Edge Compensation) algorithm, which is a modification of Maal's algorithm, is considered in more detail. The main difference of this modification is that at each processing step, not a pair of quadrature mirror filters is used, as is the case in Maal's algorithm, but only one of them. Before applying the effect of this filter, the image is processed by a Gaussian-type operator, which is a symmetric low-pass filter that smooths out brightness differences on the image plane. This contour is stored separately for each stage of image processing. A significant advantage of this algorithm is its speed compared to Maal's algorithm, this is due to the fact that only one filter is used, and not a pair of filters. Such a modification is not reflected in the qualitative and quantitative indicators of image processing, but at the same time, at each stage of decomposition, the same Gaussian operator is used to separate the contour, this procedure is repeated recursively until the image processing is completed. The characteristics of the contour selection operator may vary, but they are the same at each stage of the decomposition. The possibility of applying known methods of segmentation of medical images when using multi-scale analysis at each stage of decomposition is proposed. As a multi-scale analysis, the possibility of applying a modification of Maal's algorithm based on the classic wavelet transformation of images, i.e. MSEC (Multi Scale Edge Compensation), is considered.  This method can be used in the processing, search and segmentation of research objects on medical images.