An edge-based approach for segmentation of prostate ultrasouind images using phase symmetry

Automatic segmentation of prostate boundaries from transrectal ultrasound (TRUS) images is an important task that continues to challenge minimally-invasive surgical procedures. Several model-based and feature-based approaches have been proposed for segmentation of the prostate. In this paper, a new edge-based segmentation scheme that is based on feature phase symmetry and path cost minimization is introduced. First, the images are pre- processed with noise reduction filters. Dominant prostate edges are then detected at multiple orientations using the principle of phase symmetry, a frequency-based method that is invariant to global contrast variation. The reconstructed feature or edge map combined with an orientation map is used in a path cost minimization scheme. This scheme seeks to find the best closed contour that best match a set of edge- based rules. A number of experiments comparing the extracted contours with manually-delineated contours validated the performance of our method.