Premium
Regional differences of amyloid PET SUVR induced by spatial smoothing and the role of reference region
Author(s) -
Palombit Alessandro,
Manber Richard,
Joules Richard,
Wolz Robin
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.045178
Subject(s) - smoothing , nuclear medicine , correlation , rank correlation , mathematics , medicine , statistics , geometry
Abstract Background Spatial smoothing can be applied to imaging data from Amyloid PET (Ab‐PET) scans in clinical trials with the aim of decreasing the noise level and improve tissue delineation to the ultimate extent of easing the structural registration process. In this study, we quantify regional differences of Ab‐PET attributable to post‐reconstruction spatial smoothing. Method Amyloid PET ([18F]‐Flutemetamol, Siemens Biograph 64 mCT PET/CT scanner) and structural MRI were collected in 10 participants (3 AD, 4 MCI). Each Ab‐PET dataset was reconstructed as single saturated image (OSEM, 8 iterations, 21 subsets, PSF+TOF), separately with and without spatial smoothing (3D Gaussian kernel, FWHM = 3mm). Bias field‐corrected and skull‐stripped T1w‐MRI images were segmented with LEAP [Ledig;ISBI;2012] and separately registered to the smoothed/unsmoothed Ab‐PET volumes. Regional SUVR, defined as within‐region average with cerebellar grey matter (GC) or whole cerebellum (WHC) as reference region, was separately calculated for smoothed/unsmoothed Ab‐PET and compared between smoothed/unsmoothed Ab‐PET volumes. Result Regional SUVR was found to significantly differ between smoothed/unsmoothed (Wilcoxon signed‐rank test, p < 0.05) with both reference regions. Quantitative SUVR differences (unsmoothed – smoothed) were nonetheless regionally limited (Figure 1A) with a relative difference of ‐0.149% or +0.988% (averaged across regions/subjects) respectively with GC or WHC reference, and very small size effects with Cohen’s d = ‐0.014 (GC) and 0.046 (WHC). A strong rank correlation (Figure 1B) was found between SUVR with different smoothing levels both regarding GC (r = 0.984, p < 0.05) and WHC (r = 0.985, p < 0.05). SUVR relative differences between smoothing levels did not exceed the 5% threshold with GC (WHC similar thus omitted), differences not reaching significance (Figure 2B) if corrected for multiple‐comparisons across regions. Conclusion Smoothing‐related regional SUVR differences were found to be below the scan‐rescan physiological level [Tolboom;JNM;2009], suggesting minimal impact on the statistical analyses. Nonetheless, spatial smoothing can improve the PET‐MRI registration robustness by improving Ab‐PET tissue delineation and noise content, an essential feature for automatic processing of large cohorts’ data.