Validation of a spatially variant resolution model for small animal brain PET studies

Abstract

In small animal positron emission tomography (PET) studies, given the spatial resolution of preclinical PET scanners, quantification in small regions can be challenging. Moreover, in scans where animals are placed away from the center of the field of view (CFOV), e.g. in simultaneous scans of multiple animals, quantification accuracy can be compromised due to the loss of spatial resolution towards the edge of the FOV. Here, we implemented a spatially variant resolution model to improve quantification in small regions and to allow simultaneous scanning of multiple animals without compromising quantification accuracy. The scanner’s point spread function (PSF) was characterized across the FOV and modelled using a spatially variant and asymmetric Gaussian function. The spatially variant PSF (SVPSF) was then used for resolution modelling in the iterative reconstruction. To assess the image quality, a line source phantom in a cold and warm background, as well as mouse brain [18F]FDG scans, were performed. The SVPSF and the vendor’s maximum a posteriori (MAP3D) reconstructions produced uniform spatial resolution across the scanner FOV, but MAP3D resulted in lower spatial resolution. The line sources recovery coefficient using SVPSF was similar at the CFOV and at the edge of the FOV. In contrast, the other tested reconstructions produced lower recovery coefficient at the edge of the FOV. In mouse brain reconstructions, less spill-over from hot regions to cold regions, as well as more symmetric regional brain uptake was observed using SVPSF. The contrast in brain images was the highest using SVPSF, in mice scanned at the CFOV and off-center. Incorporation of a spatially variant resolution model for small animal brain PET improves quantification accuracy in small regions and produces consistent image spatial resolution across the FOV. Therefore, simultaneous scanning of multiple animals can benefit by using spatially variant resolution modelling.

Longitudinal preclinical evaluation of the novel radioligand [11C]CHDI-626 for PET imaging of mutant huntingtin aggregates in Huntington’s disease

“…Acquired PET data were reconstructed into 27 or 39 frames of increasing length (12 × 10 s, 3 × 20 s, 3 × 30 s, 3 × 60 s, 3 × 150 s, and 3 or 15 × 300 s) depending on the 30- or 90-min acquisition, respectively, using a list-mode iterative reconstruction with proprietary spatially variant resolution modelling with 8 iterations and 16 subsets of the 3D ordered subset expectation maximization (OSEM 3D) algorithm [ 24 ]. Frames were reconstructed on a 128 × 128 × 159 grid with 0.776 × 0.776 × 0.796 mm 3 voxels with normalization, dead time, random, decay, and CT-based attenuation corrections applied.…”

Identification of pre-synaptic density networks using SV2A PET imaging and ICA in healthy and diseased mice

“…PET data is collected and histogrammed before being reconstructed into 33 frames of increasing length (12×10s, 3×20s, 3×30s, 3×60s, 3×150s, and 9×300s). All images were reconstructed in 8 iterations and 16 subsets using the 3D ordered subset expectation maximization (OSEM 3D) algorithm utilizing a list-mode iterative reconstruction with proprietary spatially variant resolution modeling for quantitative analysis (Miranda et al, 2020). Corrections for normalization, dead time, and CT-based attenuation were applied.…”

“…In addition to human application, [ 11 C]UCB-J displayed excellent imaging characteristics in rodents and non-human primates (Bertoglio et al, 2020;Nabulsi et al, 2016;Thomsen et al, 2021). Thus, [ 11 C]UCB-J PET imaging can be applied in a preclinical setting to model neurological and neuropsychiatric disorders as demonstrated by recent studies by our and other groups (Bertoglio et al, 2022a;Bertoglio et al, 2022b;Glorie et al, 2020;Toyonaga et al, 2019) Given the brain-wide distribution of SV2A, regional analysis of SV2A PET data may be limiting the amount of information that can be obtained. In this context, data-driven approaches such as independent component analysis (ICA), a blind source separation technique, can separate the brain signal into distinct components, here defined as presynaptic density networks (pSDNs), without having any knowledge beforehand about the source signals (Bell and Sejnowski, 1995).…”

Longitudinal Preclinical Evaluation of the Novel Radioligand [11C]CHDI-626 for PET Imaging of Mutant Huntingtin Aggregates in Huntington’s Disease

“…Acquired PET data were reconstructed into 27 or 39 frames of increasing length (12x10s, 3x20s, 3x30s, 3x60s, 3x150s, and 3 or 15x300s) depending on the 30-or 90-min acquisition, respectively, using a listmode iterative reconstruction with proprietary spatially variant resolution modelling with 8 iterations and 16 subsets of the 3D ordered subset expectation maximization (OSEM 3D) algorithm [23]. Frames were reconstructed on a 128x128x159 grid with 0.776x0.776x0.796 mm 3 voxels with normalization, dead time, random, decay, and CT-based attenuation corrections applied.…”