Abstract
315
Objectives: SPECT can image a radiopharmaceutical distribution in brain according to cerebral perfusion, neurotransmitter, or cell density and therefore has many clinical applications [1]. We are developing a multi-detector multi-pinhole brain-dedicated SPECT scanner - AdaptiSPECT-C. Our selectable pinhole module technology [2] is used to adapt the collimator configuration such that each pinhole can be closed or open and image with a selectable aperture size. In this work, we explored the effects of 3 data acquisition (DAcq) schemes with different amounts of projection data overlap (i.e. multiplexing (MUX)) on activity recovery in brain perfusion imaging. Methods: The prototype design used in this work for AdaptiSPECT-C consists of 25 8” square detectors arranged in a truncated sphere with 3 rings of 10, 10, and 5 detector heads along caudal-cranial direction. Each detector is coupled with a 5-pinhole collimator consisting of one central and 4 diagonal pinholes of 2.5 mm diameter. We simulated 3 DAcq schemes as follows. Scheme #1 has 3 temporal frames: in the 1st frame, only the 1st pairs of the diagonal apertures are open; in the 2nd frame, only the 2nd pairs of the diagonal apertures are open; and in the 3rd frame, only the central apertures are open. In scheme #2, all apertures are open while in scheme #3, the 4 diagonal apertures are open, and the central apertures are not used. The digital XCAT phantom [3] was modified to simulate a brain perfusion distribution for the head size of 99th percentile male to investigate the near worst case in terms of MUX. Simulation and image reconstruction were performed by our generalizable multi-pinhole SPECT simulation and reconstruction software [4]. Defining the MUX% as the percentage of overlapped projection pixel counts to total acquired counts, schemes #1-3 have <1%, 67%, and 30% MUX, respectively, for the simulated brain phantom. Attenuation was modeled in simulation and corrected for in reconstruction; however, scatter was not included. Two Poisson noise levels of 5 M (low-statistics) and 50 M counts (high-statistics) were simulated for scheme #2. The same phantom and scan time were then simulated for schemes #1 and #3 resulted in 1.7 and 3.9 M (16.7 and 39.3 M) counts, respectively, for the low-statistics (high-statistics) simulation. A simulation using scheme #1 with 3× longer scan time was performed and considered as gold standard. Thirty-four 3D volumes of interest (VOI) were defined on the reconstructed slices. Activity Recovery % (AR%) was calculated for each VOI as the percentage of reconstructed to true activity. Results: By merging some corelated VOIs, we summarized AR% for gray and white matters in addition to a subset of subcortical regions for different DAcq schemes in Table 1. In Table 1, the values outside and inside parentheses represent the AR% for low-statistics and high-statistics simulations, respectively. The results showed that schemes #1-3 can yield similar AR% compared to the gold standard simulation. It was also demonstrated that scheme #2, despite having major amount of MUX, can generate the most accurate AR% among the 3 simulated schemes. In addition, no visible MUX-induced artefact was observed in the reconstructed images of schemes #2 and #3 and they even had images of higher visual quality compared to MUX-free scheme #1.
Conclusions: Based on a preliminary analysis of AR% of a brain perfusion phantom, we showed that in the current prototype design of AdaptiSPECT-C, opening 4 and 5 apertures per detector which allows high levels of MUX can be used to reconstruct images of AR% close to AR% in the gold standard image. Future investigations including simulating multiple noise realizations will be needed to do statistical analysis of quantification by different DAcq schemes. Research Support: National Institute of Biomedical Imaging and Bioengineering (NIBIB), National Institutes of Health, Grant No R01 EB022521.