RT Journal Article
SR Electronic
T1 Optimized Longitudinal VOIs for Amyloid PET SUVR Quantification
JF Journal of Nuclear Medicine
JO J Nucl Med
FD Society of Nuclear Medicine
SP 1923
OP 1923
VO 57
IS supplement 2
A1 Gregory Klein
A1 Joël Schaerer
A1 Florent Roche
A1 Boubakeur Belaroussi
A1 Mehul Sampat
A1 Gennan chen
A1 Joyce Suhy
YR 2016
UL http://jnm.snmjournals.org/content/57/supplement_2/1923.abstract
AB 1923Objectives PET standard uptake value ratio (SUVR) methods analyzing data in native MRI acquisition space offer the potential advantage of PET quantification with minimal smoothing. However, longitudinal amyloid analyses in native space are complicated by changing brain anatomy between time points. Intuitively, we expect that longitudinal effect size is maximized when the volumes of interest (VOIs) defined at baseline and follow up visits correspond to exactly the same brain tissue. We compare three VOI definition strategies to optimize longitudinal amyloid SUVR effect size.Methods Freesurfer (X-sectional / version 5.3) was used to obtain VOI segmentations on T1 MRI data from 385 Alzheimer’s Disease Neuroimaging Initiative (ADNI) subjects (30 with probable Alzheimer’s Dementia (AD), 200 with Mild Cognitive Impairment (MCI) and 155 Normal Controls (NC)) at two time points approximately 24 months apart. Florbetapir PET data at matching time points were registered to the MRI data in T1 native space, and composite SUVR’s were computed using a grouping of four larger cortical regions, equally weighted as described by Landau1. VOIs were defined using three strategies: 1) using baseline MRI segmentation only, 2) a cross-sectional approach independently using baseline and follow up MRI time points, each matched to the corresponding PET, 3) using propagation of baseline MRI segmentations to match MRI anatomy at follow up. Twelve potential SUVR reference regions were also evaluated including whole cerebellum, cerebellar grey (CG), corpus callosum (CC) and subcortical white matter. Longitudinal and cross sectional effect sizes for AD, MCI and NC groups were computed using Cohen’s d. Subject APOE and baseline amyloid PET burden were used to stratify the analysis and define groups of “likely decliners” in the population. Percent overlap of segmented MRI VOIs was also computed.Results The mean overlap of SUVR VOIs segmented individually at baseline and follow up was 78% and 83% for the AD and NC groups respectively. Across all reference regions, effect size was generally lowest using VOI definition strategy #1, and highest using strategy #3. The CC reference region showed the highest effect size for longitudinal change. Longitudinal effect sizes for the AD amyloid positive group using the CC reference were 1.57, 1.45 and 1.63 for VOI strategies 1, 2 and 3 respectively. Comparing across reference regions, effect size for the same group using strategy #3 ranged from a low of 0.12 (CG reference) to a high of 1.63 (CC reference).Conclusions Accurate calculations of PET longitudinal change require analysis of comparable anatomical regions between time points. Given the extent of changing brain anatomy during longitudinal studies, it appears crucial to individually match MRI to PET data time points for optimal accuracy and highest effect size. While a cross-sectional segmentation approach could be used to obtain VOIs independently from two MRI time points, results show improvement by using a longitudinal approach that propagates the baseline segmentation to its corresponding position at the follow up exam (strategy #3). Choice of reference region also greatly impacts effect size. Optimization of both VOI definition methodology and selected reference region are important for reducing required sample size in a clinical trial using longitudinal amyloid SUVR as an efficacy measure.