Visit DIX: German-Spanish-German dictionary | diccionario Alemán-Castellano-Alemán | Spanisch-Deutsch-Spanisch Wörterbuch

next up previous contents [cite] [home]
Next: Bibliography Up: Diffusion Tensor Imaging Previous: Implementation   Contents



The primary contributions of this thesis are:

The new visualization method is very useful to inspect small volumes of tensor data. It is the most precise way of displaying tensor data sets as all the information in the data is represented in the ellipsoids. Displaying the tensors as a three dimensional structure is also probably the best way to understand the properties of this data structure. Furthermore, it enables a developer to prove visually the correctness of any algorithm implemented and to check if the results of any transformation of the data has a meaningful interpretation.

Besides optimizing the code for performance, further investigation of general improvements and extensions would be interesting. Some of these ideas will be presented in the next section.

Further Work

First of all, the nonrigid registration should be expanded to fully support three dimensional data. Most functions are written to work with three dimensional data but they have not been tested. The matching process itself (Section 7.2) is only implemented for two dimensional data so this would be a starting point. Finally, an extension to three dimensions involves making sure that all functions include the voxel-dimensions correctly.

Thus, the extension to three dimensions primarily means finding a method for testing the correctness of all functions.

As it can be assumed that diffusion tensor data has no meaningful interpretation in regions of gray matter, a first step in incorporating T2 weighted data would be to mask the tensor data with the segmented white matter and use the border between gray and white matter, as a boundary condition when matching diffusion tensor data.

As any displacement of tensor data can be performed in two separate steps, (displacement of the single components and local transformation of the tensor), any displacement can be applied on tensor data. It would also be interesting to compare the results when using displacement fields derived from other then tensor data. This should be easy if the tensor data is aligned with the SPGR data. Any transformation, (rigid and nonrigid), applied on the SPGR data can also be applied on the tensor data and the corresponding local transformation of the tensors performed after the displacement of the single components.

In Section 7.3 regarding the interpolation of the data, it was mentioned that the variogram models have to be empirically selected for any given alignment. A way to find the optimal variogram model would be to select landmark points manually in a number of cases and match the corresponding points. Then the resulting displacements can be statistically analyzed to find the best variogram function for the interpolation of the displacement fields.

The visualization of the tensors as ellipsoids is only useful for small data volumes. As can be seen in the example images in Appendix B, the information provided becomes too large for an interpretation. The next step is to display the tensors as fibers, i.e. connect the tensors by following the largest eigenvalue. Fiber-tracking is a large research area; the correct identification of a path is certainly not trivial. VTK provides a class vtkHyperStreamline to connect tensors. As the images with the tensor-ellipsoids suggest, it should be possible, at least in certain regions with clear orientation of the tensors, to apply this VTK data structure to connect single tensors. The connected tensors are then visualized as a tube, which would represent a fiber. A prototype for such a function is included in the class display.h.

next up previous contents [cite] [home]
Next: Bibliography Up: Diffusion Tensor Imaging Previous: Implementation   Contents
Raimundo Sierra 2001-07-19