Summer2009:Using CUDA for stochastic tractography
- BWH: Julien de Siebenthal, Sylvain Bouix
Stochastic tractography does not provide interactive visualization so far due to its intensive computational needs.
Idea would be to visualize online paths generated in a point of interest like a fiducial. This approach would be based on the online visualization of streamline tractography done by moving a fiducial interactively.
During the summer week, we will continue our work to develop a concrete solution in investigating acceleration means based mainly on CUDA.
- Cuda was integrated through PyCuda with support for numpy arrays syntax
- Check compilation&instalation under Linux&Windows
- Implemented part of the tractography algorithm in kernel
- Tested direct call to driver api - ok
- Tested driver kernels (simple matrix operations) - ok
- Still need to complete the whole tractography algorithm in kernel mode