Segmentation of Tubular Objects by Posterior Optimization of M-reps
Rohit R. Saboo, Christina Villarruel, Edward L. Chaney, Julian G Rosenman, Stephen M. Pizer · 2007
Many organs in the human body are nearly tubular in shape, i.e., have circular cross-sections. Examples are portions of blood vessels and the colon. The method of segmentation of slab-like organs with the help of statistics on medial models has been shown to be successful for several organs. However, the medial surface of a tubular object degenerates to a curve, and the statistics of even nearly tubular objects will typically be unstable. Traditional medial models cannot handle this degeneracy. In this paper, we detail the representation, geometry and means of computing statistics on tubular medial models. We test our method on synthetic tubes with shapes sampled from the distribution of a rectum and CTs on carotid arteries and real rectums. 1.