Bi W, Weissfeld LA, Cohen AD, Klunk WE, Mathis CA, Price JC.
Exploration of the PiB positivity boundary using statistical clustering.
Human Amyloid Imaging 2011 Meeting Abstracts. 2011 Jan 15;
Objectives: Statistical clustering of PiB retention measures was performed to study retention characteristics in
subjects at the PiB+/- boundary.
Methods: PiB PET (SUVR 40-60 min) was performed for 62 control subjects (NC). Region definition and CSF
correction were MR-based. Cortical (e.g., anterior cingulate (ACG); precuneus (PRC); frontal (FRC); parietal
(PAR); lateral temporal (LTC)) and subcortical (e.g., anterior ventral striatum (AVS), thalamus, and pons) regions
were evaluated. Basic (Basic_kM) and sparse (Sparse_kM) k-means clustering were performed. Basic_kM treats
all regions with equal importance. Sparse_kM assigns regional weights that reflect regional importance in the
grouping. Two or three clusters explored PiB-, PiB intermediate (PiBint), and PiB+ groupings. Clustering results
were compared to groupings determined by a boxplot iterative outlier (IO) cut-off.
Results: Two clusters yielded 13 PiB+ NC by both methods, relative to 16 PiB+ by IO. For three clusters, Basic_kM
identified 13 PiB+, while Sparse_kM placed 4 of these in the PiBint group. A 2.5% boundary about the IO cut-off
defined 5 PiBint. Sparse_kM weights indicated 6 key regions, similar for 2 and 3 clusters: ACG, PRC: 0.21 >FRC:
0.13 >PAR: 0.11 >LTC: 0.10 >AVS: 0.09, with other regions 1.72 that was Conclusions: Statistical clustering showed objective evidence of differences in the magnitude and distribution of
As in presymptomatic controls at the PiB+/- boundary, relative to the very PiB+ or PiB-, and may prove useful as
a tool to better identify early As deposition.