Contour extraction in prostate ultrasound images using the wavelet transform and snakes

It is extremely difficult to locate the contour of the prostate in B-scan ultrasound images automatically by computer because of low resolution and high noise levels. In this paper we present a semi-automatic prostate contour extraction scheme, which is based on the wavelet transform and active-contour models, or snakes. The ultrasound image is first decomposed into edge maps at different resolutions via the wavelet transform. Seed points are found in the coarsest edge map by examining the maxima along the radial profiles which emanate from an anchor point selected manually. These seed points are used to initialize a snake, which will evolve across the edge maps at different resolutions and eventually converge to the contour of the prostate.