Using the local mode for edge detection in ultrasound images

We are investigating quantitative ultrasound texture measures as an additional source of diagnostic information for the detection of white matter damage in very preterm infants. White matter damage is a form of brain damage which leads to cerebral palsy. Ultrasound speckle properties have been shown to correlate to scatterer properties in phantoms. The disease process alters the scatterer type and density in white matter. We are enhancing speckle edges in cranial ultrasound images to determine if local speckle gradients and speckle edge densities correlate to patient outcome. Speckle edges are very diffuse and traditional edge enhancing schemes, such as Sobel, do not perform well. To capture the speckle edge detail, we use film images scanned at a very high resolution. The digitization introduces a significant noise component and does not suppress the ultrasound film grain. We present a non-linear filter to enhance the speckle edge information. The filter (DM) exploits the changes to speckle edges that result from applying local mode filtering. This technique has the advantages of maintaining edge center localization with large window sizes and performing better than Sobel for diffuse edges. In this work, we discuss the filter, its parameters, and their selection for a given application