Abstract
241332
Introduction: Multi-pinhole SPECT has the capacity to provide imaging with both high sensitivity and high resolution. Using multiple pinholes per detector can lead to overlapping projection images; this is known as multiplexing and has been shown to cause image artefacts. To avoid overlap without reducing magnification, the pinholes can be positioned such that the projections are truncated instead. Our group has previously developed an algorithm which is able to reduce multiplex-induced artefacts. Here, the impact of truncation on the performance of our de-multiplexing algorithm was considered.
Methods: This work investigated multiplexing and truncation in the context of AdaptiSPECT-C, a brain-dedicated multi-pinhole SPECT system currently under construction at the University of Arizona. AdaptiSPECT-C comprises 24 modular detectors, each fitted with an aperture plate with one central and four oblique apertures (five total apertures per detector). AdaptiSPECT-C was designed with the projections from the oblique apertures being slightly truncated to limit multiplexing. In-house, GPU-based analytic code was used to simulate the system under three aperture configurations: a configuration of AdaptiSPECT-C as it is currently being built (referred to as AS), a configuration where the oblique pinholes were translated to avoid truncation at the cost of increased multiplexing (HighMux), and one where apertures were translated to avoid multiplexing of the oblique projections at the cost of increased truncation (HighTrunc). In all cases, the magnification of the projections was not changed. Thus, there was no alteration to the system beyond that of the oblique aperture designs. All five pinholes per detector were opened for the full acquisition to maximize multiplexing.
A simulation of a spherical water phantom with 21 cm diameter (to match the system’s field of view) with a uniform distribution of I-123, and an XCAT brain phantom scaled to the 99th percentile male head size with an activity uptake mimicking an I-123-IMP perfusion study were used to assess the system under the three variations. Normalized root-mean square error (NRMSE), the structural similarity index measure (SSIM), and non-uniformity were used as metrics to assess performance. All projections were reconstructed with MLEM and utilization of our de-multiplexing algorithm.
Results: Projections and reconstructed images of the spherical phantom with the three aperture configurations are shown in Figure 1, for the original multiplexed reconstruction and up to 10 iterations of de-multiplexing. Relative to AS, the HighMux configuration resulted in a gain in sensitivity of around 8.5 %, while HighTrunc resulted in a loss of around 20 %. In all three configurations, the de-multiplexing algorithm improved NRMSE for the spherical phantom by at least 60 % for noise-free data and 30 % for data with Poisson noise. Non-uniformity was also improved through de-multiplexing in each case. The HighMux case with de-multiplexing applied gave the overall lowest NRMSE and non-uniformity for data with Poisson noise. For the brain phantom, the variation in NRMSE and SSIM was less drastic as both mux and truncation are limited even for the 99th percentile male head size (Figure 2). However, de-multiplexing still improved both metrics. For brain phantom data with Poisson noise, AS gave the best NRMSE values both prior to and following de-multiplexing. The de-multiplexed HighMux acquisition gave the overall best SSIM.
Conclusions: The use of a de-multiplexing algorithm may mean that systems can be designed with somewhat more multiplexing and less truncation, thereby improving sensitivity without degrading images. The optimal aperture configuration may depend on the activity distribution and metric of interest. The use of adaptive shuttering with these three configurations to incorporate multiplex-free frames into the acquisition is to be investigated.
Research Support: NIBIB/NIH Grant R01 EB022521
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.