Incorporating DTI Data as a Constraint in Deformation Tensor Morphometry Between T1 MR Images

Abstract

Deformation tensor morphometry provides a sensitive approach to detecting and mapping subtle volume changes in the brain from conventional high resolution T1W MRI data. However, it is limited in its ability to localize volume changes within sub-regions of uniform white matter in T1W MRI. In contrast, lower resolution DTI data provides valuable complementary microstructural information within white matter. An approach to incorporating information from DTI data into deformation tensor morphometry of conventional high resolution T1W imaging is described. A novel mutual information (MI) derived criteria is proposed, termed diffusion paired MI, using an approximation to collective many-channel MI between all images. This approximation avoids the evaluation of high dimensional joint probability distributions, but allows a combination of conventional and diffusion data in a single registration criteria. The local gradient of this measure is used to drive a viscous fluid registration between repeated DTI-MRI imaging studies. Results on example data from clinical studies of Alzheimer's disease illustrate the improved localization of tissue loss patterns within regions of white matter.

Improving Registration Using Multi-channel Diffeomorphic Demons Combined with Certainty Maps

“…multi-variate mutual information. However, the computational costs associated with estimating multi-variate mutual information makes it too cumbersome to use and calls for different assumptions or approximations to be made [11,12].…”

A Combined Manifold Learning Analysis of Shape and Appearance to Characterize Neonatal Brain Development

“…If, however, a very fine-scale co-alignment is applied, a larger proportion of the differences among the images becomes encoded in the deformations between them. In this case a study using the approach of deformation-based morphometry (DBM) becomes more appropriate [4]- [6].…”

“…Studies of pathology or searches for biomarkers may then be carried out by, for example, separating the sample into patients and healthy controls and seeking associations between the disease and the chosen feature. Studies using voxel-based morphometry (VBM) or deformation-based morphometry (DBM) form good examples of this approach [1]- [6] An alternative paradigm has begun to be applied to neuroimaging studies using methods recently developed in multivariate statistics, machine learning and pattern recognition [7]- [11]. This is based on a class of techniques developed to generate useful and low dimensional representations of high dimensional data.…”

Multimodal Image Registration by Information Fusion at Feature Level

“…Although spatially constant weights were used in the experiments, the author also suggested that a weighting function ω i (x) defined on the image domain should generate better results. In Studholme's work [4], DTI was incorporated into the mapping between T1 as an constraint. Multichannel mutual information (MI) was used to match the multimodal image.…”