[HOME]
Research Description
Registration and analysis of neuro-imaging data presents a challenging
problem due to the complex folding patterns in the human brain.
Specifically, the cortical surface of the brain can be modeled as a
highly convoluted 2D surface. Since it is non-flat, the non-euclidean
geometry of the cortex needs to be accounted for while performing
registration and subsequent signal processing of anatomical and
functional signals on the cortex. Techniques from differential geometry
offer a powerful set of tools to deal with the convoluted nature of the
cortex. We developed a surface parameterization method which computes a
2D coordinate system on the cortex which is then used to compute the
surface metric and discretize derivatives in the surface geometry. We
presented surface registration techniques to find a point to point
correspondence between two surfaces. These can be used to bring surface
signals from individual brains to a common surface. Isotropic and
anisotropic diffusion filtering methods were formulated for processing
of the cortical data. When the surface data is a point-set on the
cortex, we proposed a method to quantify its mean and variance with
respect to the surface geometry. The registration techniques presented
for surface alignment are then extended to volumes to perform full
surface and volume registration. This is done by using volumetric
harmonic mappings that extend the surface point correspondence to the
cortical brain volume. Finally, the volumetric registration is refined
by using inverse-consistent linear elastic intensity registration.
Cortical Surface Parameterization
We proposed a parameterization technique for the cortical surface based
on $p$-harmonic energy minimization. Angle and area distortion
metric were computed to evaluate the performance of this flattening
procedure. Finite element method was used for discretization. Our $p$-harmonic method results in a very fast
parameterization of high-resolution cortical surfaces and always
results in a bijective map.
Cortical Surface Registration
We present a technique that is a generalization of the popular
thin-plate spline methods from $R^n$ to a non-Euclidean surface. Our
$p$-harmonic mapping method maps each individual cortical
hemispheres to
the unit squares. We use the resulting square maps of
the cortical hemispheres to assign a coordinate system to the
cortex. We then use these coordinates to compute the metric
tensor and Christoffel symbols of the mapping. In order to
register one brain to another, we warp coordinates of one brain with
respect to another using sulcal landmarks such that the bending energy
is minimized within the true geometry of the surface. This is achieved
by solving the resulting variational problem using covariant
derivatives and thus making the warping results independent of the
coordinate system.
As an improvement over our earlier method, we also proposed an FEM based elastic mapping method that avoids the
use of an intermediate surface flattening step for landmark matching.
It incorporates the landmark registration into the parameterization
method itself. We use the Cauchy-Navier elastic equilibrium equation
for performing this matching as explained in the next section. This
approach also has the advantage that the computation cost is relatively
small and that the resulting alignment is inverse consistent.
Surface Constrained Volumetric Registration
We developed an approach to brain image registration based on harmonic
maps that combines surface based and volume based approaches producing
a volumetric alignment in which there is also a one-to-one
correspondence between points on the two cortical surfaces.
We propose a new method based on harmonic mappings for extending
the surface matching to the entire cortical volume, and present a
modified intensity alignment to
compute the final map. The resulting method, comprising the three
steps gives an inverse consistent map which is capable
of aligning both subcortical and sulcal features.
Diffusion Smoothing on Surfaces
Neuroimaging data, such as cortical thickness or neural activation, can
often be analyzed more informatively with respect to the cortical
surface rather than the entire volume of the brain. This analysis
should be carried out in the intrinsic geometry of the surface
rather than in the ambient space. We present parameterization based
numerical methods for performing isotropic and anisotropic
filtering on triangulated surface geometries. In contrast to existing
FEM-based methods for triangulated geometries, our approach accounts
for the metric of the surface. In order to discretize and
numerically compute the isotropic and anisotropic geometric
operators, we first parameterize the surface using a $p$-harmonic
mapping. We then use this parameterization as our computational
space and account for the surface metric while carrying out isotropic
and anisotropic filtering. We illustrate these methods in an
application to smoothing of mean curvature maps on the cortical
surface.
[HOME]