Abstract
313
Objectives To extract beta-amyloid (Aβ) deposition patterns based on standard uptake value ratio (SUVR) over 19 regions of interest (ROIs); To divide all subjects into two groups by automation based on cortical SUVR pattern in order to improve the clinical evaluation of early-stage cognitive impairment.
Methods 110 subjects of the ARIC-PET imaging study without clinical diagnosis of dementia received PET scans with 18F-AV-45 and MR scans in three centers respectively. In preprocessing, we compared the spatial normalization methods in Statistical Parametric Mapping (SPM) routine vs. Diffeomorphic Anatomical Registration through Exponentiated Lie Algebra (DARTEL). Clustering analysis with average-linkage hierarchical agglomerative (HA) and K-Means methods (KM) were applied to the SUVR datasets.
Results HA analysis of heterogeneous SUVR over 19 ROIs (F1, 18=152.4, p< 0.001) yielded distinct cortical and subcortical patterns. KM algorithm applied to cortical SUVR further divided all subjects into statistically positive Aβ scans (cluster 2) and negative ones (cluster 1) with no overlap (cluster 1: Mean ± SD: 1.15±0.11; cluster 2: 1.74±0.11 [p<0.001]). This method can explain 95% variability of data. There is a significant main effect of clusters × age on cortical SUVR (F2, 987=868.5, p<0.001).
Conclusions Cortical and subcortical SUVR patterns of Aβ deposition were identified. Cortical SUVR cluster analysis partitioned the subjects into statistically different groups, with higher SUVR indicating more Aβ burden in subjects of cluster 2. However, given the absence of dementia diagnoses in all subjects, these findings may imply that the Aβ deposition precedes the clinical diagnosis of dementia. In addition, the identified subcortical SUVR pattern may be linked to subcortical vascular dementia.