3D Motion Blur For PET Using GPGPU

Dóra Varnyú

Supervisor(s): László Szirmay-Kalos

Budapest University of Technology and Economics


Abstract: Computing motion blur is important not only in video games and movie rendering, but also in physical simulations where the effect of the movement has to be calculated and compensated. One such application is dynamic Positron Emission Tomography (PET), which is one of today's most important medical imaging techniques. When studying neurological disorders like epilepsy with PET, the involuntary movement of the patient during a seizure can ruin the examination results. Because of this, motion compensation should be incorporated into the reconstruction process. When the movement is rapid, accurate motion blur presents a challenge even if only 2D image space results are needed and hardware capabilities can be exploited. However, in the case of PET, the blurred result is expected in the 3D space. Blurring the motion of a 3D object in the 3D space can be done by dividing its bounding box into homogeneous voxels and tracking the path of each voxel independently. As low execution time is crucial, simplifications must be applied at the expense of accuracy. The solution proposed in this paper approximates the path of each voxel with a sequence of line segments. The segments can then be sampled by properly adapted 3D line drawing algorithms such as the Antialiased Bresenham Algorithm. Implementation is accelerated by parallel execution on GPU using the CUDA platform.
Keywords: 3D Reconstruction, Graphics Hardware, Image Processing, Scientific Visualization
Full text:
Year: 2019