Description
Diffusion Tensor Registration with Exact Finite Strain Differential (DT-REFIND) is an algorithm that registers diffusion tensor images using full tensor information. Deforming tensor images require us to interpolate AND reorient the tensors in order to be consistent with anatomy. Unfortunately, this reorientation process introduces difficulty in computing the gradient of the objective function.
In
this work, we compute an exact differential of the Finite-Strain
reorientation strategy. By utilizing the closed-form gradient and the
velocity field representation of one parameter subgroups of
diffeomorphisms, the resulting registration algorithm is diffeomorphic
and fast. Implemented within the Insight Toolkit (ITK) framework,
registration of a pair of 128x128x60 diffusion tensor volumes takes 15
minutes, which is faster than many non-linear scalar image
registration algorithms. When compared with a traditional alternative
that does not take into account the reorientation in the gradient
computation, we show that using the exact gradient achieves
significantly better registration.
Publications
- B.T.T. Yeo, T. Vercauteren, P. Fillard, J-M. Peyrat, X. Pennec, P. Golland, N. Ayache, O. Clatz. DT-REFinD: Diffusion Tensor Registration with Exact Finite-Strain Differential. IEEE Transactions on Medical Imaging, In Press, 2009. [pdf]
- B.T.T. Yeo, T. Vercauteren, P. Fillard, X. Pennec, P. Golland, N. Ayache, O. Clatz. DTI Registration with Exact Finite-Strain Differential. Proceedings of the International Symposium on Biomedical Imaging: From Nano to Macro (ISBI), 700--703, 2008. [pdf] ISBI Travel Grant
Code Release
The code is currently incorporated into the freely-available MedINRIA. MedINRIA is a graphical interface that allows the visualization and processing of medical imaging data.
Please
note that the current DT-REFinD code in MedINRIA does not perform an
initial affine registration, so for best results, you might want to
load in affine-registered volumes. In the future, MedINRIA will be
modified to allow for warp initializations.
Unfortunately, the source code is not bundled with MedINRIA. To obtain the source code, please email Olivier Clatz (olivier.clatz at inria dot fr).