An investigation of partial volume effect and partial volume correction in small animal positron emission tomography (PET) of the rat brain
ABSTRACT Partial volume correction (PVC) has been successfully applied to human PET data, where a range of methods has been used including the use of anatomical side information. The rat brain is expected to have low variability for animals of similar weight, thus making it possible to delineate volumes of interest (VOIs) on a stereotaxic atlas . The aims of this study were to investigate the magnitude of partial volume effect (PVE) in small animal PET for different regions in the rat brain and to evaluate the performance of PVC based on the geometric transfer matrix method (GTM)  using anatomical regions drawn on a stereotaxic atlas. PVE estimates in terms of activity retention in each region and spill-over between regions were calculated by convolving each region with a measured spatially invariant point spread function. PVC was tested on dynamic microPET studies of the dopaminergic D2 receptor radioligand 11C-Raclopride which were simulated using PET SORTEO, a Monte Carlo based PET simulator . The kinetics of striatum and remaining brain were simulated based on the simplified reference tissue model  using the cerebellum as the reference tissue. A significant amount of PVE is present in microPET rat brain studies with recovery of true VOI concentration being between 52% and 20%. In the simulated 11C-Raclopride study the uncorrected time activity curves showed up to 55% reduction in measured activity concentration and a bias in binding potential of up to −36%. Good activity recovery and improvement of binding potential estimation was achieved with PVC (−0.26% to −4.36% bias). We conclude that PVE has a substantial influence on rat brain studies and PVC should be used to improve quantitative accuracy. PVC using the adapted GTM method shows promising results.
- [Show abstract] [Hide abstract]
ABSTRACT: The aim of this study was to evaluate the potential of anatomy-based reconstruction, using microCT information, to improve quantitative accuracy in multiple-pinhole SPECT. Multiple-pinhole SPECT and microCT images were acquired with the Micro Deluxe Phantom using both hot and cold rod inserts. The phantoms were filled with 3.7 MBq/ml of (99m)Tc. To improve microCT contrast, the phantoms were also filled with contrast agent. Emission images were reconstructed using a one-step-late (OSL) modification of the ordered subsets expectation maximization (OSEM) algorithm for incorporation of microCT information, to encourage smoothing within but not across boundaries. To allow quantification, the OSL OSEM algorithm takes into account imperfect camera motion, collimator response, angular variation of the sensitivity, intrinsic camera resolution, attenuation and scatter. For comparison, the emission images were also reconstructed by OSEM using post-reconstruction filtering and by OSL OSEM using a quadratic prior and an edge-preserving prior. In each rod of the phantoms the recovery coefficient (RC), defined as measured divided by the true activity concentration, was expressed as a function of the noise. Different noise levels were obtained by varying the amount of spatial filtering during or after reconstruction and by the use of binominal deviates. Compared to conventional OSEM using post-reconstruction filtering and compared to OSL OSEM using a quadratic prior, our study demonstrated that the use of anatomical information during reconstruction significantly improved the quantitative accuracy in both cold and hot rods with a diameter larger than or equal to 2.4 mm. When compared to the edge-preserving prior, the anatomical prior performs significantly better for hot rods with a diameter ≥ 2.4 mm. For the 4.0-mm hot rods for example, the RC averaged over the different noise levels was 0.67 ± 0.02 when multiple-pinhole SPECT images were reconstructed using anatomical information, compared to 0.54 ± 0.08, 0.60 ± 0.04 and 0.64 ± 0.02 when OSEM in combination with a post-reconstruction filter, OSL OSEM using a quadratic prior and OSL OSEM using a median root prior was used, respectively. For the 4.0-mm cold rods, the RC averaged over the different noise levels was 0.61 ± 0.03 when the multiple-pinhole SPECT images were reconstructed using anatomical information, compared to 0.54 ± 0.07, 0.53 ± 0.08 and 0.60 ± 0.03 when OSEM in combination with a post-reconstruction filter, OSL OSEM using a quadratic prior and OSL OSEM using a median root prior was used, respectively. Anatomy-based reconstruction using microCT information has the potential to improve quantitative accuracy in multiple-pinhole SPECT.European Journal of Nuclear Medicine 09/2010; 38(1):153-65. DOI:10.1007/s00259-010-1627-6
- [Show abstract] [Hide abstract]
ABSTRACT: Quantitative accuracy in rat brain PET studies is reduced by partial volume effect. We investigated the performance of partial volume correction (PVC) in a realistic situation where activity is also taken up in the head and spills into the brain. The PVC approaches studied include the region-based geometric transfer matrix (GTM) method and voxel-based iterative deconvolution (reblurred Van Cittert and Richardson-Lucy). 8 realizations of dynamic rat brain PET studies of 11C-Raclopride with a binding potential BPND=3 in the striatum were simulated with the Monte Carlo simulator PET SORTEO. Synthetic time activity curves (TACs) were assigned to the striatum, cerebellum, remaining brain and head regions outside the brain of a rat head phantom. Different sized volumes of interest (VOIs) were sampled ranging from the full anatomical region to smaller VOIs containing only voxels with at least 50%, 70% or 90% of the maximum activity. BPND was calculated for the striatum using the simplified reference tissue model with the cerebellum as the reference tissue. Without PVC the accuracy of BPND was very low for all VOI sizes with biases between −44.7% and −20.9%. PVC using the GTM method was only accurate for the smallest 90% VOI with a bias of −7.7% but the standard deviation increased to 4.2% compared to less than 1% for the larger VOIs. Good accuracy was achieved for both iterative deconvolution methods using the 50% VOI (bias less than 8%) with standard deviations of less than 1.8%. Thus, in the presence of activity uptake outside the brain, iterative deconvolution methods outperform the GTM method. We are currently implementing PVC with a spatially variant PSF to better compensate for non-uniformities of spatial resolution away from the centre of the field of view.01/2011; DOI:10.1109/NSSMIC.2011.6153722