GPU-Based Fast Iterative Reconstruction of Fully 3-D PET Sinograms

Abstract

Abstract-This work presents a graphics processing unit (GPU)-based implementation of a fully 3-D PET iterative reconstruction code, FIRST (Fast Iterative Reconstruction Software for [PET] Tomography), which was developed by our group. We describe the main steps followed to convert the FIRST code (which can run on several CPUs using the message passing interface [MPI] protocol) into a code where the main time-consuming parts of the reconstruction process (forward and backward projection) are massively parallelized on a GPU. Our objective was to obtain significant acceleration of the reconstruction without compromising the image quality or the flexibility of the CPU implementation. Therefore, we implemented a GPU version using an abstraction layer for the GPU, namely, CUDA C. The code reconstructs images from sinogram data, and with the same System Response Matrix obtained from Monte Carlo simulations than the CPU version. The use of memory 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 original code running in a single core of a high-end CPU, with no loss of accuracy.

Fast Patch-Based Pseudo-CT Synthesis from T1-Weighted MR Images for PET/MR Attenuation Correction in Brain Studies

“…We assigned a standard 18 F-FDG PET activity to each tissue according to relative values (gray matter, 4.0; white matter and rest of soft tissue, 1.0; cerebrospinal fluid and bone, 0.0) (24,25), obtaining the groundtruth PET maps. Then, these ground-truth maps were projected with the 3D ordered-subsets expectation maximization software (26), assuming the geometry, parameters, and sinogram format of the Biograph mMR scanner (27). The system response matrix used in this case assumed a uniform gaussian point-spread function of 4 mm in full width at half maximum in the whole field of view.…”

“…Both the attenuated projection data and the attenuation-free data were reconstructed with 3D ordered-subsets expectation maximization software (26) adapted to the geometry and sinogram size of the mMR scanner. We used 30 iterations and 3 subsets, and the reconstructed images consisted of 201 · 201 · 129 voxels of 2 · 2 · 2 mm each.…”

The Effect of Mouth Motion on the Attenuation Correction in Neurological PET Studies

“…The "ideal" emission sinogram was "attenuated" using the image with the mouth completely closed and then it was corrected with an erroneous attenuation projection, obtained from the image with the mouth relaxed. This final sinogram was reconstructed with the OSEM algorithm (3 iterations 20 subsets) [13] and the differences between the images reconstructed with the proper attenuation correction (which corresponds to the image obtained from the emissions sinogram) were studied.…”

Medical image processing on the GPU – Past, present and future

“…Barker et al (2009) accelerated the most computationally demanding steps of MOLAR PET, which includes motion compensation to improve the image quality. Herraiz et al (2011) made a CUDA implementation for iterative reconstruction of sinogram PET, which is much more common in commercial scanners than listmode PET.…”