Deriving resolution-compensated PET images from the GTM partial volume correction method

Introduction: PET suffers from spatial resolution limitations and the ensuing partial volume effects (PVE) greatly impact accurate tracer quantification. Partial volume correction (PVC) methods include region-based (ROI) and voxel-based (image) approaches. While PET has commonly involved regional (ROI) assessment of activity concentrations due to individual voxel noise, it can be desirable to obtain actual images obtained with PVC for further inspection and analysis.

Methods: While the region-based geometric transfer matrix (GTM) method [1] is widely used for PVC, it has sometimes been toned down for only providing ROI values, and not being able to provide actual images corrected for PVE. At the same time, images corrected for PVE using the GTM-PVC method can be produced following some extra computing steps. We applied our GTM-PVC method [1] to a cohort of cognitively normal volunteers (N=10) and to patients diagnosed with Alzheimer’s disease (N=10), who underwent amyloid load assessment using two different PET tracers ([¹¹C]-PiB, and [¹⁸F]-florbetapir). Dynamic PET data were acquired on a GE Advance^® scanner, and the last 4 frames from the 70min studies, assumed to be in a steady state, were averaged and used for analysis. MRI data were used to build individual tracer models (Fig. 1) derived from a cloud-based service [2] using multi-atlas segmentation approach [3]. PET simulations were carried out to derive GTM data for each of the 31 regions (VOI) labeled in the tracer model (Fig. 1). Subject-specific MRI-derived ROI templates were built upon thresholding (50%) each volume-of-interest (VOI)’s axial component of the regional spread function [4]. Correction maps derived from the GTM-PVC data were then generated using the concept of apparent recovery coefficient [5] [AR1] and applied to the original PET data to yield PVC images (GTM-PIX, Fig. 2). After applying the GTM-PVC method, corrected ROI values were then compared with the values obtained from application of the same ROI template to the GTM-based pixel (GTM-PIX) images.

Results: Regional values obtained from the GTM-PIX method were compared with regional values obtained after GTM-PVC using the same ROI template. Regional activity concentrations from the GTM-PIX PVC method were found to be in excellent agreement with regional values obtained with the GTM-PVC method, typically showing close to one-on-one correlation, with R² values typically > 0.99 (Fig. 3).

Conclusions: We show how easily it is to derive images from the GTM correction method, not requiring any additional input, and that can be very easily implemented following application of the GTM PVC method. This method provides results consistent with those obtained with the GTM-PVC after sampling the GTM-PIX images with the same ROI templates. Acknowledgements: This work was supported in part by R21 grant no. AG056142 from the National Institute on Aging. References: [1] Rousset et al., J Nuc Med (1998); [2] Mori, S. et al., Computing in Science & Engineering (2016); [3] Wang, H. et al., Inf Process Med Imaging (2013); [4] Rousset et al., J Nucl Med (2008); [5] Rousset et al., Comp Med Imag Graph (1993)