GPU-based fast iterative reconstruction of fully 3-D PET sinograms

摘要

This work presents a graphics processing unit (GPU)-based implementation of a fully 3-D PET iterative reconstructioncode, FIRST (Fast Iterative Reconstruction Software for [PET] Tomography),which was developed by our group. We describe themain steps followed to convert the FIRST code (which can run onseveral CPUs using the message passing interface [MPI] protocol)into a code where the main time-consuming parts of the reconstructionprocess (forward and backward projection) are massively parallelizedon a GPU. Our objective was to obtain significant accelerationof the reconstruction without compromising the imagequality or the flexibility of the CPU implementation. Therefore,we implemented a GPU version using an abstraction layer for theGPU, namely, CUDA C. The code reconstructs images from sinogramdata, and with the same System Response Matrix obtainedfrom Monte Carlo simulations than the CPU version. The use ofmemory was optimized to ensure good performance in the GPU.The code was adapted for the VrPET small-animal PET scanner.The CUDA version is more than 70 times faster than the originalcode running in a single core of a high-end CPU, with no loss ofaccuracy.