The objective of this proposal is to test two hypotheses of the imaging characteristics of positron emission tomography (PET): (1) that substantial improvements in reconstruction accuracy can be obtained for whole-body PET systems by super-sampling voxels during acquisition through the near-continuous motion of the patient bed along a direction intentionally - but carefully and accurately - misaligned with respect to the scanner's axial direction;and (2) that super-sampling will allow resolution improvement with iterative algorithms using resolution-recovery techniques, beyond what those algorithms can achieve without super-sampling. Super-resolution through wobbling is a technique historically used in early brain PET systems to improve the sampling, which would typically improve resolution and reduce artifacts. With the introduction of block detectors and whole-body systems, wobbling disappeared because its mechanical cumbersomeness and added expense were not justified since the impact on reconstruction was negligible. This is because the block detectors greatly reduced the crystal size, reducing the importance of crystal size on overall system resolution in comparison with other effects such as acolinearity and signal distortions in readout. Thus, the sampling improved disproportionately to the resolution. Two significant changes have occurred since wobbling was removed from scanners: (1) 3D acquisitions greatly increased the count statistics, which supports the accurate reconstruction of smaller features;and (2) resolution-recovery iterative algorithms have become commonplace. We believe that improving sampling will result in increased accuracy and resolution of modern reconstructions while artifacts will be reduced because resolutions are once again becoming better in comparison to the sampling. We propose a novel method for increasing sampling, including axial sampling, that does not require additional hardware, making the solution possibly attractive to vendors for future systems and as low-cost upgrades to existing systems, if the hypotheses of this proposal are confirmed. The method introduces a small angle in the bed alignment combined with the use of existing bed motors to super-sample the object, requiring no additional scan time. When the bed moves axially, a small transaxial shift results.
The specific aims of this proposal include: (i) determining through simulation the optimal stepping pattern and bed-alignment angle;(ii) modifying existing reconstruction software to incorporate the bed-position information;and (iii) experimentally test the method on a TOF research scanner, LaPET, that is in our laboratory.
These aims will allow us to fully address the hypotheses of this proposal.
This project develops a new technique for improving sampling in PET by deliberately aligning the bed at a small, nonzero angle to the axial direction of the scanner. As the bed moves through the scanner the voxels positions relative to the scanner's lines of response (LORs) change, improving sampling, analogous to the early brain PET systems that wobbled the scanners. The improved sampling is likely to result in more accurate reconstructions with better resolution, particularly as resolution-recovery algorithms improve in their system modeling, which will allow for higher quality images in clinical PET. .
| Li, Yusheng; Defrise, Michel; Matej, Samuel et al. (2016) Fourier rebinning and consistency equations for time-of-flight PET planograms. Inverse Probl 32: |
| Li, Yusheng; Matej, Samuel; Metzler, Scott D (2016) A unified Fourier theory for time-of-flight PET data. Phys Med Biol 61:601-24 |
| Li, Yusheng; Defrise, Michel; Metzler, Scott D et al. (2015) Transmission-less attenuation estimation from time-of-flight PET histo-images using consistency equations. Phys Med Biol 60:6563-83 |
| Li, Yusheng; Matej, Samuel; Karp, Joel S et al. (2015) LOR-interleaving image reconstruction for PET imaging with fractional-crystal collimation. Phys Med Biol 60:647-70 |
| Li, Yusheng; Matej, Samuel; Metzler, Scott D (2014) Image reconstructions from super-sampled data sets with resolution modeling in PET imaging. Med Phys 41:121912 |
