Abstract
A new speckle reduction method, i.e., Laplacianpyramid-based nonlinear diffusion (LPND), is proposed for medicalultrasound imaging. With this method, speckle is removedby nonlinear diffusion filtering of bandpass ultrasound imagesin Laplacian pyramid domain. For nonlinear diffusion in eachpyramid layer, a gradient threshold is automatically determinedby a variation of median absolute deviation (MAD) estimator.The performance of the proposed LPND method has been comparedwith that of other speckle reduction methods, includingthe recently proposed speckle reducing anisotropic diffusion(SRAD) and nonlinear coherent diffusion (NCD). In simulationand phantom studies, an average gain of 1.55 dB and 1.34 dBin contrast-to-noise ratio was obtained compared to SRAD andNCD, respectively. The visual comparison of despeckled in vivoultrasound images from liver and carotid artery shows that theproposed LPND method could effectively preserve edges anddetailed structures while thoroughly suppressing speckle. Thesepreliminary results indicate that the proposed speckle reductionmethod could improve image quality and the visibility of smallstructures and fine details in medical ultrasound imaging.
Index Terms—Laplacian pyramid, multiscale analysis, nonlineardiffusion, speckle reduction, ultrasound imaging
I. INTRODUCTION
ULTRASOUND imaging has several advantages over othermedical imaging modalities (e.g., X-rays, computedtomography, and magnetic resonance). It is safe, relatively lowcost, and allows real-time imaging. However, the diagnosticusefulness of ultrasound imaging is at times limited due to itslow image quality. One of the main reasons for this low imagequality is the presence of signal-dependent noise known asspeckle. Speckle is a granular pattern formed due to constructiveand destructive coherent interferences of backscatteredechoes from the scatterers that are typically much smallerthan the spatial resolution (i.e., wavelength of an ultrasoundwave) of medical ultrasound systems [1]. The speckle pattern depends on the structure of the imaged tissue and variousimaging parameters, e.g., the frequency and geometry of anultrasound transducer. Thus, two images captured under thesame condition will show an identical speckle pattern. Becauseof its dependence on the microstructure of tissue parenchyma,speckle is often used in diagnosis, such as diffuse liver diseases[2]. However, speckle typically shows up as noise since itreduces image contrast and obscures image details. Also, itaffects human interpretation of the acquired ultrasound imagesand degrades the speed and accuracy of ultrasound imageprocessing tasks.Generally speaking, there are two main purposes for specklereduction in medical ultrasound imaging. First, speckle reductioncan aid human interpretation of ultrasound images. Second,speckle reduction is a preprocessing step for many ultrasoundimage processing tasks such as segmentation and registration.When applying a speckle reduction technique as an aid for visualdiagnosis, we need to keep in mind that certain specklecontains diagnostic information and should be retained. Also,some important details may be smeared or lost when performingspeckle reduction. In some cases, clinicians prefer an originalimage to a despeckled image, because the original image containsmore diagnostic information. From this point of view, thedespeckled image should be considered a complement to theoriginal image, and not a replacement. When speckle reductionis applied as a preprocessing step for segmentation or registration,any speckle can be considered noise without differentiation.In these applications, speckle prevents image segmentationand registration techniques from generating optimal results.Speckle reduction makes an ultrasound image cleaner withclearer boundaries, and thus significantly improves the speedand accuracy of automatic or semiautomatic image segmentationand registration techniques.Several speckle reduction techniques based on compoundingand postacquisition image processing techniques have been proposed[3]–[23]. With the compounding techniques, a series ofultrasound images of the same target are acquired from differentscan directions [3], with different transducer frequencies [4],or under different strains [5]. Those images are then averagedto form a composite image. While the compounding methodscan improve target detectability, they suffer from degraded spatialresolution and increased system complexity. On the otherhand, the postacquisition image processing techniques do notneed many hardware modifications. The postacquisition techniquesfor speckle reduction can be classified into two categoriesingle scale spatial filtering and multiscale methods.



Download full report
http://hompi.sogang.ac.kr/bmslab/publica...uction.pdf