Approach to boundary detection in ultrasound imaging

Ultrasound imaging systems pose unique challenges for standard edge detection algorithms, because the boundaries between regions of interest in an ultrasonogram are typically bright streaks between similar intensity regions, rather than demarcations between regions of differing contrast. This paper presents an approach to boundary identification in ultrasound images that sidesteps this problem by the use of operators that work parallel to the edges rather than perpendicular, as most edge detection procedures do. These operators, called `stick,' are line segments short enough that they can locally approximate the edges in the image, but long enough that their projections onto background noise are insignificant. The algorithm produces a new image by plotting at every point the local projection of the original image onto the stick whose orientation relative to the image maximizes the projected value. The result is an image with reduced speckle noise and enhanced edges in which boundaries between different regions can be more easily detected. This approach has been applied to real-world distance estimation research; processed images indicate distances which correlate highly with physical measurements of the actual tissue.