A Non-Local Low-Rank Framework for Ultrasound Speckle Reduction

‘Speckle’ noise refers to the granular patterns that occur in ultrasound images due to wave interference. Speckle removal can greatly improve the visibility of the underlying structures in an ultrasound image and enhance subsequent post-processing. We present a novel framework for speckle removal based on low-rank non-local filtering. Our approach works by first computing a guidance image that assists in the selection of candidate patches for non-local filtering in the face of significant speckle noise. The candidate patches are further refined using a low-rank minimization estimated using a truncated weighted nuclear norm (TWNN) and structured sparsity. We show that the proposed filtering framework produces results that outperform state-of-the-art methods both qualitatively and quantitatively. This framework also provides better segmentation results when used for pre-processing ultrasound images. 1. Motivation and Related Work Medical ultrasound is a widely used noninvasive imaging modality that can reveal internal anatomic structures. Ultrasound makes use of a transducer to emit ultra-highfrequency sound waves, which change direction when a reflective surface is encountered. Careful timing of the emitted sound signal and its observed echo is used to determine the anatomical structures. One drawback of ultrasound imaging is the noise that results from wave interference when the scattered waves constructively and destructively combine to produce the black and white ‘speckle’ pattern characteristic of ultrasound images [3, 14]. Figure 1 shows a typical ultrasound image and the granular pattern appearance of the speckle noise. The presence of speckle noise lowers the overall image quality and makes the interpretation of ultrasound images challenging for nonspecialists [22, 30]. Speckle noise can also adversely affect the identification of normal and pathological tissues by trained specialists [8, 19]. Furthermore, it lowers the accuracy of computer-aided diagnosis [8] and adversely affects subsequent image processing tasks such as Figure 1: Top A typical clinical ultrasound image corrupted with speckle noise. Bottom The despeckled and speckle noise layers recovered by our proposed method. segmentation [2, 5]. Over the last two decades there have been a number of methods proposed to reduce speckle noise. A number of wavelet-based methods have been proposed to decompose the ultrasound image into frequency subbands and then use various strategies to filter wavelet coefficients associated with speckle noise (see [7] for an overview of wavelet-based methods). However, these frequency domain approaches tend to oversmooth the image details by filtering excessive frequencies, or produce ringing artifacts due to removal of incorrect bands [32]. Another popular strategy for speckle removal are local image filtering methods. Among these methods, the most successful ones are those based on anisotropic diffusion (e.g., [19, 8, 31]) and the bilateral filter (e.g., [2]). While local filters are successful for speckle reduction, their performance suffers in the presence of strong noise, which corrupts the correlations between neighboring pixels [10]. In addition to local filtering, non-local filtering methods have also been proposed. Methods such as non-local means (NLM) [5, 32, 29] leverage the entire image by finding similar patches in a larger neighborhood of a target pixel. The collection of patches is then used to filter the target pixel.