Graduation Project at the LKEB
GPU-based Medical Image Registration
Supervisor: Marius Staring, PhD (LKEB, dept. Radiology), Leiden University Medical Center (LUMC)
Background
In the modern operating room intra-operatively acquired imaging data is used amongst others for navigation, the guidance of surgery, and radiotherapy. These intra- operative images are then spatially related to pre-operative images from a treatment plan, based on CT or MRI. The spatial relation between the data can be found by an image registration procedure, which is modeled as a mathematical optimization problem. Currently, this procedure is very slow, rendering it unusable in this setting.
Image registration as an optimization problem.
Research Objectives
In this project you will exploit parallelism to accelerate (high-dimensional) nonrigid 3D image registration procedures, using GPUs. Research questions include the identification of components suitable for parallel computation, and the creation of a design to optimally exploit parallelism, on the GPU using OpenCL. Challenges include hardware intrinsics (memory limitations, communication overhead, etc).
Research will be performed using the open source image registration package elastix (http://elastix.isi.uu.nl), where an initial OpenCL framework is already available. As an additional design constraint the modularity of the software should be kept intact, which benefits usability but will influence design and ultimately raw performance.
The development of algorithms and software require (GPU) programming skills.
For more information:
Marius Staring
m.staring@lumc.nl, 071 52 62137