Detection of cancer in animal tissues: a wavelet approach

Abstract

Considering that the biospeckle laser is a dynamic interferometric phenomenon adopted as a tool to monitor changes in biological samples and that the temporal variation of speckle pattern depend on the activity level of the sample surface illuminated, this work proposes to analyse the time-varying scale-mixing matrix. Using two-dimensional scale-mixing wavelet transform several descriptive summaries varying on time are derived. These descriptors are signature of image regularity and fractality useful in tissue classification. In this work we propose to verify the behavior of the energy-flux between the scales, considering a set of 128 images to classifying cancer areas in images of an anaplastic mammary carcinoma in a female canine and in images of skin cancer in a cat obtained over time. The time-varying spectral slopes applied in the analysis of dissimilarities of tissues allowed to note that healthy area descriptors have lower values than cancer area descriptors, resulting in higher Hurst exponents. By using scaling properties of tissue images, we have captured information contained in the background tissue of images which is not utilized when only considering traditional morphological analysis.