A comparison of three different filters for speckle reduction of Doppler spectra

This paper is concerned with speckle suppression in Discrete Fourier Transform based Doppler signals by means of digital image processing. The Doppler spectrum is treated as a greyscale image, and three different noise smoothing algorithms are applied to it. These are the Double Window Modified Trimmed Mean filter, which is nonlinear, and Lee's and the directional filter, which are adaptive in the sense that the smoothing performed by them at each point of the image is controlled by a local image measure. In order to evaluate the performance of the filters, they were applied to a variety of regular waveforms obtained from a physiological flow phantom. Ensemble averaging of a large number of unfiltered spectra was used as the "gold standard" in the evaluation, i.e., as the output of an ideal filter which reveals the exact nature of the underlying Doppler spectrum after speckle has been eliminated. Comparison of the "gold standard" with the ensemble averaged filtered data allowed the noise reduction, bias and distortion of the maximum frequency envelope introduced by filtering to be examined. Overall, the best performance was offered by the directional filter whose action was controlled by the combination of the local edge content and the slope of the least-squares-fit line passing through the data points along each particular direction.