Evaluation of PVC methods on head‐to‐head FTP and MK6240

Abstract

AbstractBackgroundMK6240 and Flortaucipir (FTP) are tau PET tracers with differing off‐target signal. Due to partial volume effects (PVE), this off‐target may spill into neighboring cortical regions. We applied different partial volume correction (PVC) methods to MK6240 and FTP images acquired in the same participants within two months to evaluate the performance of the methods within and between tracers.Method15 subjects (Table1) were scanned using MK6240 (SUVR calculated 70‐90min and 90‐110min) and FTP (80‐100min) on a Siemens Biograph mCT and smoothed to 8mm3 resolution. T1 MRIs were segmented using FreeSurfer 7.1. Five different PVC methods were tested: 2 compartment, 3 compartment, Van‐Cittert (VC), geometric transfer matrix (GTM) and region‐based voxelwise (RBV). Different ROI configurations were used when applying GTM/RBV. To determine how well the PVCed data fit the original data, PVCed images were smoothed back to the 8mm3 resolution and compared to the original image within SPM‐segmented gray matter, white matter, CSF, skull and meninges (Figure1).ResultVC fit the original data best after smoothing (Figure2), but had the highest correlation with the original data (Figure3c), meaning VC does not correct the data as much as the other methods and therefore does not address the PVE issue. 2 and 3 compartment fit the gray and white matter well (Figure2) while showing low R2 between the original and PVC data (Figure3). However, these methods assume zero signal outside the cortex, resulting in poor fit of extracortical areas and leading to over‐amplification of the cortical signal. GTM and RBV perform well for both tracers in all 5 regions. Their correlations between original and PVC data are between 2/3 compartment and VC, indicating these methods correct the data. PVC improves the correlation between FTP and MK6240; the highest correlation between tracers is achieved when the same PVC method is applied to both tracers (r2=0.9 for 2 and 3 compartment, r2=0.89 for GTM, r2=0.88 for RBV, r2=0.84 for VC, r2=0.80 for original).ConclusionGTM/RBV assume non‐zero signal outside the cortex resulting in a better fit of the original data, correct the data more than VC, and improve the correlation between MK6240 and FTP.