Session S81.1

Motion Estimation in X-Ray Rotational Angiography Using a 3-D Deformable Coronary Tree Model

A Bousse*, J Zhou, G Yang, J-J Bellanger, C Toumoulin

Université de Rennes
Rennes, France

This paper deals with the reconstruction of a sequence of coronary 3-D skeletons at different cardiac phases from a single X-ray rotational angiography sequence. This reconstruction is performed via the deformation of an initial coronary tree 3-D model using gated X-ray rotational tomographic 2-D projections, where the 2-D centerlines are assumed to be extracted. This way to proceed allows to track in time each point of the initial 3-D skeleton, and to build a sequence of functions that map a 3-D skeleton of any phase to the initial tree. The estimated motion is intended to be used for motion compensation in a dynamic tomographic reconstruction process.
The deformation process consists in a minimization of a functional that is the sum of a data fidelity term and of an internal energy term. The first term is a functional that shrinks as the 3-D skeleton matches the projections corresponding to the considered cardiac phase. The internal energy controls the deformation of the 3-D skeleton. Several aspects can be controlled: proximity of neighboring points, spacial regularity of the motion, fidelity to the initial coronary tree.
Once the sequence of coronary trees is reconstructed, a set of B-spline motion functions is estimated such that each point of each reconstructed skeleton is approximately mapped to its source point in the original tree. Spline coefficients are computed through the minimization of a quadratic cost function that includes fidelity and regularity terms.
We tested our method with 20 3-D centerlines that were previously extracted from a 3-D dynamic sequence acquired on a 64-slice GE LightSpeed CT coronary angiography. This sequence included 20 volumes reconstructed at every 5% of the RR interval. The 3-D centerlines were projected to their corresponding projection planes in order to obtain 80 projections over 4 cardiac cycles (20 projections per cycle). The initial 3-D skeleton and these 80 2-D projections were then used to estimate the 3-D centerlines at each phase. Wrong centerlines were also added to test the robustness of our method. Results appeared to be satisfying, in term of robustness and convergence.

(Abstract Control Number: 258)