Expoentes de Hurst em imagens interferométricas de um carcinoma mamário

Abstract

This research aimed to develop new applications for image analysis based on the decimated and non-decimated discrete wavelet transform. We estimated and verified the behavior of the Hurst exponents (directional) using the wavelet transform in 128 interferometric images, obtained at regular intervals, of a canine anaplastic mammary carcinoma. The first procedure consists of extracting parts of the images that portray cancerous and healthy tissues and applying the multiresolution analysis methodology at four levels of resolution using the Daubechies wavelet with eight null moments. The second procedure was based on estimating Hurst’s exponents in the 256 sub-images, using different methods to decrease the correlation of wavelet coefficients and the effect of possible outliers, thus conducting a temporal analysis of these exponents over time. We verified the non-overlapping series distributions in at least one direction when comparing sub-images of cancerous and healthy tissue, concluding that the temporal analysis of Hurst exponents may differentiate them. By applying this methodology, it will be possible to improve the quality of the information obtained from images of cancerous tissues by recognizing patterns in the Hurst exponent that can be used along with a clinical analysis in the tumor classification or detection processes.