Shape Recovery using Deformable B-Spline Models.

As part of my Ph.D. research I developed a shape recovery system using deformable B-Spline surfaces. I used it to extract part of the cerebral vascular system that was imaged using magnetic resonance angiography (MRA). This research was some of earliest work into creating patient specific vascular models that could then be used for surgical planning or for creating physical phantoms that could used for improving imaging techniques.

The system employed 6 steps:

Segmentation using a seeded region growing technique.

Centerline construction using a skeletonization technique.

Graph construction using a simple centerline following technique.

Model construction that used a series templated deformable B-Spline surfaces.

An image based deformation process to fully recover the shape of the vessels.

A constrained optimized process to obtain/maintain geometric (G1) continuity in the B-Spline surfaces after the deformation process.

More details of this work can be found in:

Allen R. Sanderson. "Shape Recovery of Volume Data With Deformable B-Spline Models," Ph.D. Dissertation, University of Utah 1996. [ PDF ]

ABSTRACT

In many fields today such as radiology, images of interesting structures are obtained. From these images the physician or scientist attempts to make decisions using a va riety of techniques. The existing techniques for representing and analyzing particular structures include volume rendering, and surface renderings from contours, free-form sur faces and geometric primitives. Several of these techniques are inadequate for accurate representations and studying changes in the structure over time. Further, some of these techniques have large data requirements that prevent interactive viewing.

It is believed that if the structures of interest can be extracted from the image back ground, viewed and analyzed in an interactive setting, more accurate decisions can be made. The research described in this dissertation explores a new technique for shape recovery with deformable models using B-spline surfaces. The current literature shows that there have been many successful attempts to create deformable models but always at a loss in shape information and/or continuity. To overcome these limitations, we show that template models each having a unique topology can be developed and the template models can be joined without a loss in smoothness at their boundaries. Further, new techniques have been developed to extract and relate image data to these deformable models.

This research work was conducted under the guidance of Elaine Cohen, Tom Henderson, and Dennis L. Parker while at the University of Utah.