A new model for the recovery of cylindrical structures from medical image data

Abstract

We introduce a novel analytic model formulation for recovering cylindrical structures (e.g., blood vessels) from segmented 3-D medical image data. Unlike all previous formulations, our model is capable of describing a cylinder with an arbitrary spine (a space curve based on cubic B-splines) and arbitrary cross section which is guaranteed to be orthogonal to the spine. Given this expressiveness, we are able to provide a second order continuous approximation to the centerline of nearly any tubular object. This information may be used for such tasks as a reformatting of the original image data in order to visually detect stenoses or aneurysms. In addition, the cross-section parameter values of our model may aid in automatically isolating these regions. We maintain a relatively simple cross-section function to make this detection straightforward (note that any cross-section function is possible). To describe fine detail in the data, we employ local finite element deformations from the model surface. Thus we are able to recover gross geometric approximations as well as quantify characteristics of the object such as its surface area. We apply our model to the recovery of both a healthy and diseased aorta from segmented CT acquisitions.