This talk will discuss efforts to develop several image analysis
approaches aimed at the recovery of left ventricular (LV) deformation
from image sequences acquired from one of several different modalities.
Initially, a Bayesian a-posteriori estimation approach will be described
that requires the accurate segmentation of the endocardial and
epicardial surfaces in each image frame. Shape (curvature) features are
computed from these surfaces and are used as deformation tracking tokens
to estimate a sparse set of local displacements. These displacements are
then used as external (data-based) forces that drive an anisotropic,
linear elastic, biomechanical model implemented using the finite element
method (FEM). The Bayesian strategy permits tradeoffs between data and
model smoothing/interpolation, ultimately estimating a dense set of
myocardial strains between pairs of image time frames. Results from
canine and human data will be shown derived from both 4D magnetic
resonance images (MRI) and echocardiographic images. In vivo validation
results using implanted markers and sonomicrometers using canine
datasets will also be shown.
We will also discuss work performed that uses features derived directly
from image intensity data (rather than segmented surfaces) and extended
free form deformation strategies, within the robust point matching
framework (RPM). This strategy reduces the dependency on precise surface
segmentation and the difficulties sometimes encountered in trying to
mesh complex geometries. In addition, the approach employs deterministic
annealing for improved optimization capture range. Further work will be
then shown that takes some of the principles related to
intensity-derived feature tracking and RPM and re-introduces a
biomechanical model using the boundary element method (BEM). Again in
vivo results will be shown using both MRI and ultrasound image sequences.
Finally, we will present a planned integrated framework for LV
deformation analysis that integrates segmentation and deformation
estimation, incorporates both shape features and mid-wall features based
on echocardiographic speckle tracking again using a biomechanical
framework that will include temporal regularization. Initial speckle
tracking results related to this will be presented. This effort is joint
work with Xenios Papademetris, Ping Yan, Ning Lin and Yun Zhu.
Audio (MP3 File, Podcast Ready)
Copyright. All Rights Reserved.