Accurate fully 3D list-mode PET reconstruction on multi-GPUs
Résumé
List-mode based image reconstruction is continuously gaining ground in Positron Emission Tomography (PET) imaging. Such reconstruction has many benefits as the preservation of the temporal sampling of the acquired data, which facilitates the study and correction of dynamic processes, as well as maintaining the highest spatial sampling available for a given detector geometry. The centerpiece of this reconstruction is the ray projector which unfortunately also represents the most time-consuming part of the reconstruction process. In this study, we propose a new projector that incorporates detector corrections including geometric and detector scatter effects to improve the image accuracy. As the computation burden is a main obstacle to obtain such reconstruction, we propose a powerful method to implement the reconstruction on multiple graphics processing units (Multi-GPUs). Results showed that our projector provides reconstructed images with a high accuracy (low noise, high contrast and resolution), with a GPU implementation speedup of x44 compared to a CPU one.