(EN) Fiber Tracking

Fiber tracking

Responsible researcher: Adiel Mittmann
Last updated: 2010-01-14

Results of fiber tracking

Fiber tracking is the name of a process that finds the fiber bundles in a patient’s brain. The picture to the right shows typical results. Fiber tracking is especially important for neurosurgeons, because by looking at the wiring of a brain he or she is able to find the best way to execute a surgery, causing as little damage to the brain as possible. One important feature of fiber tracking that makes it a valuable tool for preoperative planning is that it is non-invasive, since it is an MRI-based technique—more precisely, it is based on DT-MRI, which is short for diffusion tensor magnetic resonance imaging.

Our group has developed a software framework for fiber tracking and other DT-MRI applications. Currently, it implements the method of streamline fiber tracking, which works by finding possible fibers from a set of seed points. Each seed point spawns one trajectory, whose points are determined by following the orientation of water diffusion. By looking carefully at the enlarged version of the pictures in this page, one may notice that they are made up of hundreds of tubes; each one of those was calculated from a seed point.

GPU-based fiber tracking

Results of fiber tracking

Fiber tracking is a rather complex process that requires a great amount of computation to be performed. In order to make it run faster, our group has implemented the process of fiber tracking for GPUs (graphics processing units), which are the processors of regular video cards. This means that actual fiber tracking is taking place on the video card, not on the CPU.

Although it is harder to write programs to be run on the GPU, there is a good reason to do so in the case of fiber tracking: speed. There are two main features of GPUs that make them an attractive platform:

  • A parallel environment: while today’s CPUs possess but a few cores, typical GPUs have dozens of them, which makes it possible to execute a great quantity of threads simultaneously. In the case of fiber tracking, the GPU is able to work on the calculation of many trajectories at the same time.
  • Fast floating-point operations: because of their origin in graphics processing, GPUs are able to execute calculations involving floating-point numbers very fast, which is useful for fiber tracking, since it is a process that requires interpolation of tensors and integration of vector fields.

We have shown how fiber tracking can be run on GPUs by using the Cg language, which was specially designed for the writing of GPU programs. Cg programs can be executed on a wide range of NVIDIA video cards and is a good choice when compatibility with older hardware is desired.

The performance gain achieved by running fiber tracking on GPUs varies from system to system, but we have obtained gains of up to 40 times when using high-end GPUs. Thus, a process that used to take 5 minutes can now be executed in under 8 seconds.

Real-time fiber tracking

The speed achieved by executing fiber tracking on the GPU made us consider an idea that was not previously feasible: real-time fiber tracking on regular computers. With real-time fiber tracking, the volume of interest can be freely dragged around the brain while fiber tracts are being constantly recomputed. This way, one can quickly explore a brain by moving the volume of interest and changing parameters, such as minimum fractional anisotropy and minimum mean diffusivity, without giving up on interactivity.

Real-time fiber tracking was only possible because our laboratory is also developing a medical workstation, which, in this application, is responsible for all user interaction and, most importantly, for drawing the resulting trajectories on the screen. Please check our video on YouTube, which contains an exploratory session demonstrating real-time fiber tracking.

Researchers

  • Adiel Mittmann: main fiber tracking researcher.
  • Tiago H. C. Nobrega: workstation/computer graphics researcher.
  • Eros Comunello: project vice-coordinator.
  • Aldo von Wangenheim: project coordinator.

Publications

  • A. Mittmann, T. H. C. Nobrega, E. Comunello, J. P. O. Pinto, P. R. Dellani, P. Stoeter, A. von Wangenheim, Performing Real-Time Interactive Fiber Tracking, Journal of Digital Imaging, 2010. (In Press)
  • A. Mittmann, Real-Time Fiber Tracking through Graphics Processing Units, 2009. (Master’s thesis)
  • A. Mittmann, M. A. R. Dantas, A. von Wangenheim, Design and Implementation of Brain Fiber Tracking for GPUs and PC Clusters, in: 21st International Symposium on Computer Architecture and High Performance Computing, 2009.
  • A. Mittmann, E. Comunello, A. von Wangenheim, Diffusion tensor fiber tracking on graphics processing units, Computerized Medical Imaging and Graphics, 2008.
  • A. Mittmann, D. D. Abdala, L. Coser, A. von Wangenheim, A framework for DT-MRI applications, in: X Congresso Brasileiro de Informática em Saúde (CBIS), 2006.

Sobre Adiel Mittmann

Possui graduação em Bacharelado em Ciências da Computação pela Universidade Federal de Santa Catarina (2005) , mestrado em Ciências da Computação pela Universidade Federal de Santa Catarina (2009) e curso-tecnico-profissionalizante pela SENAI - Departamento Regional de Santa Catarina (2001) . Tem experiência na área de Ciência da Computação. Atuando principalmente nos seguintes temas: Tractografia, GPU, DT-MRI, Computação de alto desempenho.