TY - JOUR T1 - Validations for MR-based partial volume correction for brain PET imaging JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 1930 LP - 1930 VL - 57 IS - supplement 2 AU - Keisuke Matsubara AU - Masanobu Ibaraki AU - Tetsuya Maeda AU - Toshibumi Kinoshita Y1 - 2016/05/01 UR - http://jnm.snmjournals.org/content/57/supplement_2/1930.abstract N2 - 1930Objectives Several methods to correct partial volume effect (PVE), based on magnetic resonance (MR) images, have been proposed and utilized widely in PET studies. However, procedures for partial volume correction (PVC) have not achieved a sufficient consensus on how we should perform PVC for brain PET studies yet. We aimed to reveal effects of data processing procedures in MR-based PVC: 1) PVC algorithms; 2) MR segmentation; and 3) post-reconstruction smoothing.Methods Brain [18F]FDG PET scans with five healthy controls (age: 61 - 75 years) were performed at 30 - 60 minutes after intravenous injection of [18F]FDG. Stardard uptake values (SUV) were calculated from PET images reconstructed with filtered back projection (FBP). To reveal the effect of post-reconstruction smoothing, the reconstructed images were smoothed with Gaussian kernel with 0 - 12 mm full-width half-maximum (FWHM), corresponding with 5 - 13 mm FWHM in final spatial resolution. The SUV maps were corrected for PVE with two MR-based voxelwise PVC algorithms: modified Müller-Gärtner’s method (mMG) (Muller-Gartner et al., 1992); and region-based voxelwise (RBV) (Thomas et al., 2011). First, the SUV values and their intra-region variability (%CoV = S.D. / mean [asterisk] 100), as an index for uniformity at each region of interest (ROI), were compared among two PVC algorithms and without PVC. Second, SUV values with PVC based on different MR segmentations were compared among three MR segmentation methods: SPM12, FAST on FSL and FreeSurfer. Finally, we compared PVE-corrected SUV values among with post-reconstruction smoothing with the 0 - 12 mm FWHM kernels. ROIs for comparison among PVC algorithms and post-smoothing were determined automatically with FreeSurfer. For comparison among the segmentations, ROIs based on automated anatomical labeling (AAL) template masked with gray matter (GM) mask, estimated with each MR segmentation method, were utilized. The results with 6 mm FWHM post-reconstruction smoothing were compared to reveal the effects of PVC algorithms and MR segmentations.Results SUV values at GM regions corrected with both RBV and mMG were significantly higher than ones without PVC (p < 0.05, paired t-test): for example, 4.8 ± 1.2 without PVC; 8.2 ± 2.2 with mMG and RBV at whole cerebral cortex. Intra-region variability was improved with RBV in all cortical regions: for example, %CoV: 19.5% (without PVC) and 11.4% (with RBV).With mMG, the intra-region variability increased in any regions, including occipital cortex. SUV maps corrected with RBV were visually different among the three segmentation methods: the highest SUV with FreeSurfer and the lowest with SPM12. The corrected SUV values in cortical regions were significantly different among three segmentation methods (p < 0.05, ANOVA), except for in sensory motor cortex. The inter-method differences were 24% at maximum [FreeSurfer - SPM12 in occipital cortex]. Few differences of SUV to without post-reconstruction smoothing (< 10%) were observed with post-reconstruction smoothing with the kernels smaller than 6 mm FWHM. Large differences (> 10%) between without and with post-smoothing were observed in some regions with larger kernels than 8 mm FWHM.Conclusions We revealed three findings: 1) PVC with RBV can improve intra-region uniformity; 2) SUV values with PVC depends on choice of MR segmentation method; and 3) excessive post-reconstruction smoothing induces bias of outcomes with PVC. Our findings show evidences to build consensus how we should perform in brain PET studies. ER -