Filtrado wavelet en estudios planares óseos de Medicina Nuclear

Abstract

Planar scintigraphy images present problems for the detection of small lesions, due to high noise contamination levels and low spatial resolution. In this work, an algorithm is introduced in order to reduce Poisson noise in these kind of images, by using Anscombe transformation followed by wavelet filtering. Material and methodsSimulated, noise-free lesions were inserted on bone images from real patients. Each image was contaminated with Poisson noise by using various noise intensities. Anscombe transformation was applied with the purpose of treating Poisson noise as gaussian noise. The images were then filtered into the wavelet domain using different filters. Finally, the quality of the images was assessed by using objective measurements, such as signal to noise ratio gain, normalised mean-squared error, and structural similarity index over regions of interest. ResultsIt was found that by applying Anscombe transformation before wavelet filtering, the resulting image quality was better in all cases, with a significant reduction of the noise levels (p=.015), with no noticeable deterioratio in the spatial resolution. ConclusionThe filtering process using wavelets coif3 with 5 decomposition levels, bior 3.5 with 5 levels, and db2 with 4 levels, showed the best results in our experiments. These results were better than those obtained using traditional filters.