A high resolution and high sensitivity Prism-PET brain scanner with non-cylindrical decagon geometry

1136 Introduction: We have designed a high spatial resolution and high sensitivity brain PET scanner based on our Prism-PET detector modules. High sensitivity is achieved by noncylindrical decagon shaped geometry which fits the ellipse shape of the human brain and improves solid angle coverage. The parallax error (PE) introduced by large oblique angles of detector array can be suppressed by using the Depth-of-interaction (DOI) readout provided by Prism-PET detectors, and therefore achieving a uniform 1-mm level FWHM spatial resolution across the scanner’s field of view (FOV). In this work, the simulation result of the Prism-PET brain scanner is presented. Methods: GEANT4 application for tomographic emission (GATE) was used to simulate both Prism-PET brain scanner and Biograph Vision (BV, Siemens Healthineers, USA) for comparison purposes. Detailed simulation parameters are listed in figure 1(b). The spatial resolution performance evaluation was implemented according to NEMA(National Electrical Manufacturers Association)-NU4-2008 and NEMA- NU2-2012 standard using 0.3mm diameter 18F sphere source embedded in 10 mm plastic cube. Image quality was evaluated by using 6-size hot-spot phantom (Figure 1(g)) with rod diameter of 4mm, 3.2mm 2.4mm, 1.6mm, 1.2mm, and 1mm at both center and edge, and a 3D voxelized Zubal brain phantom with 256x256x128 matrix size and 1.1x1.1x1.4 mm3 voxels. 18F source was used in rod phantom with 250 kBq/cc uniform activity. Figure 1(c) demonstrates the process flow for image reconstruction. To simulate the inter-crystal scattering (ICS) recovery ability enabled by the Prism-PET detector, the raw list mode data from the GATE simulation was firstly performed the ICS correction based on the hits file obtained from the GATE simulation. Then, each DOI-line of response was projected onto the first front DOI bin of the neighbored crystal to form the DOI corrected list-mode data (DOI-rebinning). The list-mode data were reconstructed with CASToR (Customizable and Advanced Software for Tomographic Reconstruction) using OSEM (10 iteration:8 subsets) with 0.4 x 0.4 x 0.4 mm3 voxels and PSF modeling. The normalization and attenuation correction were performed by feeding the normalization data and attenuation image into the CASToR. Results: The highest absolute sensitivity of the Prism-PET brain scanner reached 14% and was improved to 17.3% by extending the axial length into 14 rings. According to the simulation data, only around 53% of events were ICS free; 21% were scattered and detected within 3 neighbor crystals; and the rest were detected by the crystal that is 3 crystals away from the original interaction site. This indicated that ICS is one of the major challenges for ultra-high-resolution blocks with a sub-millimeter level crystal size. The geometry of decagon arranged modules increases the fraction of gamma photons that hit the crystal surface with an oblique angle, thus Prism-PET suffers from a high level of PE even at the center of FOV. Original hot spots phantom of Prism-PET showed the impact of PE on image quality at both center and edge of the FOV of Prism-PET. With both DOI-rebinning and ICS recovery, Prism-PET can resolve hot spots down to 1 mm in diameter, in comparison, BV can only resolve 3.2 mm spots. Conclusions: The Prism-PET brain scanner can achieve a very high sensitivity due to the decagon geometry which maximizes the solid angle. Moreover, the PE can be corrected using DOI readout, and thus achieving a uniformly distributed high spatial resolution across the entire FOV of the Prism-PET brain scanner. Acknowledgments: We thank the Center for Biotechnology (CFB) at Stony Brook University for their financial support through the NIH REACH program.