Estimação dos expoentes de Hurst com utilização da mediana e do estimador trimean em imagens interferométricas de um carcinoma mamário

Abstract

The research aimed to develop new applications for image analysis based on the non-decimated discrete wavelet transform. The behavior of (directional) Hurst exponents were verified, over time, using the non-decimated wavelet transform in 128 interferometric images of a canine anaplastic mammary carcinoma, obtained in regular time intervals by means of the dynamic biospeckle method. In order to achieve the goal, sub-images that depict only carcinogenic tissues were extracted, as well as sub-images that depict only healthy tissues, and applied with the multiresolution analysis methodology with four levels of resolution using the Daubechies wavelet with eight null moments. So as to estimate, the mean of the squares of the wavelet coefficients, the MEDLA method and the trimean estimator (GTME) were used, and their effects on the series of Hurst exponents were also verified. We concluded that the temporal analysis of the Hurst exponents is an alternative tool to distinguish between images of the healthy tissue and the carcinogenic one. The data suggest that the best method of estimation of the Hurst exponent was the MEDLA method as it produces less time variability.