TISSUE CHARACTERIZATION BY ULTRASOUND SPECKLE

Abstract

Ultrasound speckle, which is caused by the interference of the backscattered ultrasoundsignal, is considered to degrade image quality and is therefore unwanted in images. A variety ofresearch involving the statistical analysis of speckle has been performed and methods have thusbeen developed to suppress or remove ultrasound speckle from images. Recent evidence has however shown that speckle contains useful information and is part of the signal. Research istherefore currently needed to use speckle to assist recognition of formerly unrecognizable objects.

It is often difficult to differentiate various soft tissues in ultrasound images due to the similarities intheir physical (elastic) properties. Since speckle texture in ultrasound images is believed to containinformation about internal tissue structures, local area texture analysis was performed on speckleimages. Pixel-based features were obtained from multi-scale resolution analysis and pixel value runlength analysis. The features were analyzed for their ability to discriminate tissues. After featureselection, multivariate classification techniques were used to differentiate beef muscle from porkmuscle in animal ultrasound images. Morphological operations were applied to produce bettersegmentation. The neural network algorithm classified 84.3% of pixels into the correct class, whileusing only the ultrasound signal in a similar algorithm correctly classified a mere 71.7%. Theseresults indicate that the developed algorithms can be used in soft tissue differentiation and can bepotentially used in medical ultrasound image to help with diagnosis.