Parametric voxelwise analysis is a commonly used tool in neuroimaging, as it allows for identification of regions of effects in the absence of a strong a-priori regional hypothesis by comparing each voxel of the brain independently. Due to the inherent imprecision of single voxel measurements, spatial smoothing is performed to increase the signal-to-noise ratio of single-voxel estimates. In addition, smoothing compensates for imprecisions in anatomical registration, and allows for the use of cluster-based statistical thresholding. Smoothing has traditionally been applied in three dimensions, without taking the tissue types of surrounding voxels into account. This procedure may be suitable for subcortical structures, but is problematic for cortical regions for which grey matter often constitutes only a small proportion of the smoothed signal. New methods have been developed for cortical analysis in which voxels are sampled to a surface, and smoothing is restricted to neighbouring regions along the cortical grey matter in two dimensions. This procedure has recently been shown to decrease intersubject variability and bias of PET data. The aim of this study was to compare the variability, bias and test-retest reliability of volumetric and surface-based methods as they are applied in practice. Fifteen healthy young males were each measured twice using the dopamine D1 receptor radioligand [11C]SCH-23390, and analyses were performed at the level of individual voxels and vertices within the cortex. We found that surface-based methods yielded higher BPND values, lower coefficient of variation, less bias, better reliability and more precise estimates of parametric binding. All in all, these results suggest that surface-based methods exhibit superior performance to volumetric approaches for voxelwise analysis of PET data, and we advocate for their use when a ROI-based analysis is not appropriate.
Copyright © 2017 The Authors. Published by Elsevier Inc. All rights reserved.