Speckle detection for 3D ultrasound

Positioning accuracy of 3D ultrasound slice planes can be enhanced by speckle decorrelation of the B-scans. A suitable method requires accurate identification of zones of non-specular reflection in the image, a feature of parenchymal tissue. This has traditionally been achieved with statistics such as signal-to-noise ratios and measures of autocorrelation curves. Co-occurrence matrices and structural approaches have also been applied to ultrasound images. In this paper we assess the performance of these methods when only B-scan display data is available. We also report improved results with in vivo images using a proposed structural algorithm that detects speckle kernels. All detection algorithms yielded detection error rates between 28% and 38%, suggesting that user intervention may be required for clinically useful performance.