Analysis of multi-dimensional tagged MR cardiac images with deformable models.

摘要

In this thesis, a thorough study of object segmentation and motion tracking by deformable models in cardiac magnetic resonance imaging (MRI) is presented. We developed automatic methods to extract 2D left-ventricular contours by deformable curves, 3D left-ventricular walls by deformable surfaces, and 3D interior material points of the heart by 3D deformable solids. The overall system tracks the motion and computes the strain of all myocardial material points by a 4D spatio-temporal B-Spline model from tagged MRI. Finally, a 5D statistical analysis characterizing the spatio-temporal change of strain across normal porcine hearts is presented.;We have employed the Chamfer distance potential for fitting models to data with conjugate gradient descent optimization. The system was validated using a cardiac motion simulator to determine the accuracy of reconstructed motion fields, strain, as well as tag intersections. The 4D system has been tested while adjusting the B-Spline order and internal energy weighting coefficients. The speed of the system and the effect of tag frequency have also been determined. The system has been tested on a variety of porcine and human data sets.