Modeling Complex Axon Fiber Architecture

The objective is to develop tools for estimating multiple as well as single intra-voxel diffusion profiles based on brain Diffusion Weighted (DW) MR images. Nowadays, this is an important research area due to the fact that approximately one third of white matter voxels contain more than a single fiber. Our aim is to provide low-requirement methods that can be applied to clinical studies.

description

Our work is based on the Multi Diffusion Tensor (MDT) fitting framework. We are developing and improving tools which perform an MDT fitting by using a prefixed basis of DW signals. The advantages of this model are that the solution does not require high computational burden and does not use excessive high-quality data.

Current activities include

• Performance evaluation

Analysis of the computed solutions in order to characterize the quality expected for clinical data.

• Orientation improvement

The improvement of multi-tensor estimation by taking into account local white-matter diffusion features as well as single DT information.

dw-data spatial regularization

Local multi-Tensor information can be corrupted by noise in the DW signal or by an insufficient number of DW measurements. In order to improve the local estimations we perform robust spatial data integration over the multi-tensor fields. This task must be done carefully, in such a way that real information is not removed, and false orientations are not introduced.

Current activities include

• Spatial analysis

Characterization of homogeneous white-matter regions in order to improve existing spatial integration methods.

images

single DT classical DT fitting. Likely fiber-crossing sites present a plate shape. Click on the image to enlarge it.

multi-DT fitting by DBFs Our proposed multi-tensor estimation is very competitive with respect to state-of-the-art methods. For in-vivo human data, our method estimates 1-, 2- and 3-bundle fiber orientations at places which are congruent with prior anatomical knowledge. Click on the image to enlarge it.

non-regularized fiber orientations The local DBF orientations could be prone to error because of a low signal-to-noise ratio, the complex fiber structure, and/or a reduced number of measurements. Click on the image to enlarge it.

regularized orientations Our experiments show that the robust spatial data regularization significantly improves the local estimations. The Robust Regularized multi-tensor field is correctly aligned while preserving the subtle axonal structures. Compare with the image on the left. Click on the image to enlarge it.

bibliography