Integration of GPGPU Methods into a PET Reconstruction System

S. Schaetz and M. Kucera (Germany)


PET, GPGPU, CUDA, OSEM, medical solution, image reconstruction


Positron emission tomography (PET) is a medical imaging modality that allows the detection of tissue with particular properties or the observation of metabolic processes in living organisms over time. It plays a vital role in cancer diagnosis and heart examination and represents a promis ing method for early diagnosis of dementia. With the advancement of scanner technology as well as an ever increasing demand for higher quality images and higher patient throughput the image reconstruction system can be come a bottleneck of a PET system. To meet this demand for more computing power in the reconstruction system a novel way to speed up the image reconstruction process for modern PET systems is presented. A common graphics processing device in conjunction with the Compute Unified Device Architecture (CUDA) is used to speed up existing algorithms for the CPU. Projectors for full 3D reconstruc tion are ported to the CUDA device and new algorithms are implemented where necessary. The main focus of this port is to speed up the calculation and to maintain the numerical accuracy of the result. On the basis of the CPU reconstruction system a new system that uses GPU projectors is implemented. With this modification a two-fold speed up is achieved compared to the highly optimized CPU implementation running on 4 CPU cores. Different optimization methods are explored and applied where suitable. The results calculated by GPU and CPU projectors are numerically identical allowing them to be interchanged. The system is validated and the algorithms are integrated into the reconstruction system of Siemens Healthcare Molecu lar Imaging PET scanners.

Important Links:

Go Back