Yujin Jang1,
Helen Hong1, Hak Jong Lee2, Sung Il Hwang2
1Division of Multimedia Engineering,
Seoul Women's University, Seoul, Korea, Republic of; 2Department
of Radiology, Seoul National University Hospital of Bundang, Seongnam-si,
Korea, Republic of
To
segment the prostate in MR images with a poor tissue contrast and shape
variation, we propose a reliable and reproducible segmentation method using a
prior knowledge of shape, geometry and gradient information. The prostate
surface is generated by 3D active shape model using adaptive density profile
and multiresolution technique. To prevent holes from occurring by the
convergence of the surface shape on the local optima, the hole is eliminated
by 3D shape correction using geometry information. In the apex of the
prostate which has a large anatomical variation, the boundary is refined by
2D contour correction using gradient information.