Session P84.2

Parametric Modeling of the Beating Heart with Respiratory Motion Extracted from MR Images

G Pons Moll, G Crosas Cano, G Tadmor, RS MacLeod,
B Rosenhahn, DH Brooks*

Northeastern University
Boston, MA, USA

In atrial fibrillation ablation procedures on-line measurement of catheter position, e.g. through magneto-tracking devices, is often displayed to the clinician against a static anatomy resulting from pre-procedure scans. However the catheter is localized in laboratory rather than anatomical coordinates, and the heart is moving due to both contraction/dilation and respiratory motion. As a result both small-scale (beating) and large-scale (respirator) inaccuracies are introduced into the on-line representation. As part of a larger project to improve delivery of ablation, we have developed parametric models for use in animating a static three-dimensional pre-procedure anatomical data to enable representation of the on-going dynamics. The steps for making the heart "beat" start with automatically localizing the heart using the temporal variance of the heart region. Extraction of the heart border is performed by minimization of an appropriately tailored energy function. Finally, a 3D tensor array is constructed from all time surfaces using cylindrical coordinates and time and space consistency of the segmentation is enforced by fitting an m-variate tensor smoothing spline to the final 3D array. The result is an efficient parameterization of the moving surface over both space and time. The steps for making the heart move due to respiration are only partially complete. We begin with transferring to a spherical coordinate representation of the surface to allow more efficient interpolation of the surface. We next developed a procedure to extract an arbitrary 2D planar slice of this model given a description of the desired plane. We are currently working on extracting a 4D model of respiratory motion from a sequence of 3D slices (2D in space plus time), from which automatic slice-to-slice temporal registration must be extracted. The approach involves parameterizing motion of curves representing anatomical landmarks and enforcing consistency in the cross-slice direction (for instance, the coronal direction for saggital slices). The next step will be to use the respiratory model parameterization to drive the bulk heart motion as itself parameterized by translation and 2-axis rotation. The final step will be to combine the respiratory and contraction/dilation models together.

(Abstract Control Number: 222)