© IC-SCCE STATISTICAL VERSUS WAVELET-BASED DE-SPECKLING TECHNIQUES FOR ENHANCING MEDICAL ULTRASOUND IMAGES

Abstract. This study performs a comparative evaluation of two statistical and three wavelet-based speckle suppression filters as means of improving the diagnostic potential of medical ultrasound images. Three sets of ultrasonic images (25 liver, 32 kidney, and 18 breast) were selected by connecting the video output of an HDI-3000 ATL digital ultrasound system to a Screen Machine II frame grabber, using 512×512×8 image resolution. Two statistical filters, the adaptive weighted median filter (AWMF) and adaptive speckle suppression filter (ASSF), and three wavelet thresholding speckle reduction filters, the soft wavelet thresholding (SWTF), the hard wavelet thresholding (HWTF) and the Garrote wavelet thresholding (GWTF) filters, were developed in C++. For speckle embedding, simulated images were degraded by Gaussian additive noise. The performance of each filter was assessed by means of mean square error (MSE) and signal-to-MSE, as well as by expert physician’s evaluation. AWMF achieved the highest noise suppression and SWTF scored the highest signal-to-noise ratio. Furthermore, SWTF has been proved superior by the physician’s blind-review evaluation, while at the same time required the minimum processing time, rendering its routine plausible for clinical application.