Despeckling of ultrasound images of bone fracture using multiple filtering algorithms

Abstract

a b s t r a c tUltrasound images are popularly known to contain speckle noise that degrades the quality of the images for good and fast interpretation in many areas of medicine, especially for bone fracture detection. This necessitates the need for robust despeckling techniques for clinical practice. Therefore, a study was carried out to reduce speckle using filtering algorithms such as Wiener, Average, Median, Anisotropic Diffusion and Wavelets. This paper discusses the level of improvement obtained through these filtering algorithms using the peak signalto-noise ratio (PSNR) as a measurement tool. The results of our work presented in this paper suggest that the combination of Daubechies-Wiener which we call as a hybrid technique with Anisotropic Diffusion, gave the best performance, which is a new contribution in this field. This despeckling algorithm can be further developed and evaluated at a larger scale.

Speckle Noise Reduction in B-Mode Echocardiographic Images: A comparison

“…(1) where gðx; yÞ is noise free image, fðx; yÞ is the acquired noisy image, and nðx; yÞ is the multiplicative noise; x and y are the variables indicating the spatial locations [5,7,9]. The process of converting multiplicative noise into additive can be carried out by applying log [11], Kaun [12], Frost [13] [ 2] À TTE image in PLAX and PSAX CED [14] [ 14] À CT bone, [9] À carotid artery [11][12][13] À SAR images, [ …”

“…The log operation decouples the multiplicative components into independent individual noise-free and noise components [2,5,6,20]. Filters such as DPAD [17], generalized likelihood estimation method (GLM) [28], NLM [39], PPB [40], and anisotropic TV (ATV) [19] are implemented using the multiplicative noise model and all other filters [6][7][8][9][20][21][22][23][24][25][26]34,41,42,46,47] are implemented using log transformation of the input image.…”

“…Signal to noise ratio [2,9] Computing of the level of speckle before and faster processing PSNR Peak signal to noise ratio [5][6][7]9,10] Measurement of objective difference between two images MSE Mean square error [2,6,9,34] Mean difference between original and despeckled image r Correlation coefficient [34] Measurement of closeness between original and processed image RMSE Root mean square error [9,10] Square root of squared error average over a window AD Average difference [9,10] Mean difference divided by the size of the image SC Structural content [9,10] Measurement of similarity, for identical images the value should be 1 LMSE Laplacian mean square error [9,10] Obtain the edge features of the image MD Maximum difference [9,10] Maximum difference between original and processed image Err3…”

“…Zhang et al [5] have compared the performance of nonlocal means (NLM) filters with local adaptive, anisotropic, hybrid, and multiscale techniques on US images of breast. Elamvazuthi et al [7] analysed six filters US of bone fracture. Loizou et al [9] and Loizou and Pattichis [10] compared applications of 10 denoising filters for suppression of speckle noise in carotid artery.…”

A novel fuzzy filter for speckle noise removal

“…In [29], a nonlinear diffusion filter with anisotropic behavior was used for denoising speckled images. A hybrid technique with anisotropic diffusion was used in [30]. A combination of the Daubechies-Wiener methods was used in the hybrid technique for speckle removal from ultrasound images.…”