Ultra-high spatial and time of flight resolution brain PET reconstruction

Introduction: Despite recent advances in PET detectors and image reconstruction algorithms, the spatial resolution of clinical PET systems is in the range of 3 mm for best dedicated brain PET systems, and the time of flight (TOF) coincidence timing resolution (CTR) is limited to 200 ps FWHM. To address these limitations, we are developing a dedicated brain PET scanner expected to achieve ultrahigh TOF resolution (~75 ps FWHM), and spatial resolution (∼1.2 mm), substantially improving upon existing scanners. We also take advantage of the faster timing resolution to revolutionize the conventional cylindrical geometry of current PET scanners. In this work, we present results of a Monte Carlo simulation study demonstrating the performance of the proposed brain PET system and how this compares to existing commercial PET scanners. Methods: The proposed PET system is composed of 1x1x10 mm3 LSO crystals arranged in 8x4 modules, each with 32x32 channels and a crystal-crystal pitch of 1.6 mm to match the photodetectors (Figure 1.A). The scanner radius is 17 cm and the height is 22 cm. Ultra-high TOF resolution (75 ps FWHM) is achieved by using multiplexed SiPMs [1] with improved photon detection efficiency (PDE) in combination with FastIC ASIC chips [2, 3]. This ground-breaking TOF resolution enables the use of smaller scintillators which will enhance the intrinsic spatial resolution and minimize resolution-degrading parallax effects. To demonstrate the performance of the proposed system, acquisitions of a high-resolution brain phantom (BigBrain) using the proposed scanner were simulated using GATE. Similar simulations were performed for a state-of-the-art clinical PET scanner (Siemens Biograph Vision with 200 ps FWHM TOF resolution). For both scanners, 140 million recorded coincidences were collected. Results: The simulated list-mode data were reconstructed using an in-house reconstruction engine which implements the ML-EM algorithm with a TOF-enabled distance-driven projector. Images were reconstructed with an isotropic pixel size of 0.8 mm. The number of iterations was selected to match the noise level between the two scanners. For the brain phantom, the imaging model of the proposed scanner included point-spread function (PSF) modeling simultaneously in the image and projection domains. Image domain PSF was modeled by a spatially invariant Gaussian blurring kernel with FWHM of 1 mm. Projection domain PSF modeling was implemented via a distance-dependent in-plane Gaussian kernel. Gaussian kernels for different radial positions were estimated from GATE simulations of point sources at different distances from the isocenter, followed by a quadratic fit of the point response FWHM as a function of the radial distance. Reconstructions of our model of Siemens Biograph scanner were simply post-processed by a Gaussian filter with FWHM of 1.5 mm. Conclusions: A novel ultra-high TOF resolution PET system was evaluated and compared to an existing state-of-the-art TOF PET scanner, showing substantially improved spatial resolution, thanks to the ultra-high TOF resolution and advanced system model. These results suggest that the proposed system may allow the visualization of small brain structures which are typically not visible with existing systems. This could enable new opportunities for neurobiology research in aging and related fields.