Landau SM, Price JC, Jagust WJ, Mathis CA.
Reliability of longitudinal PIB: How do data processing methods influence detection of change over time?.
Human Amyloid Imaging 2011 Meeting Abstracts. 2011 Jan 15;
The ability to accurately measure amyloid longitudinally is critical for making inferences about how amyloid
changes relate to clinical progression or response to amyloid-modifying drugs. Furthermore, these changes are
likely to be small, so distinguishing them from processing-related artifacts is particularly challenging. Recent
Pittsburgh Compound B (PiB) PET studies have reported test-retest variability of 1-6% for normal older controls
and 3-10% for Alzheimer's patients (Tolboom et al 2009; Aalto et al 2009), indicating that grealh amyloid changes
of around 5-10% may be difficult to distinguish from artifacts of image processing methods, such as smoothing,
intensity scaling, and spatial normalization.
We investigated the influence of imaging processing methods on detection of longitudinal amyloid changes by
comparing the results of two processing streams: (Method 1) defining ROIs and a cerebellar reference region by
spatially normalizing each PIB scan to an MCI template, and (Method 2) defining ROIs and a cerebellar reference
region in native space using regional parcellation with Freesurfer software. Both processing streams were fully
automated, and no partial volume correction was applied. PIB-PET imaging data was acquired longitudinally
through the Alzheimer's Disease Neuroimaging Initiative (ADNI). 103 subjects (20% normal controls, 20%
Alzheimer's patients, 60% MCI patients) had baseline PiB scans, 80% had 1-year follow-up scans, and 40% had
2-year follow-up scans.
Annual change in the non-scaled cerebellum was similar for Alzheimer's, MCI, and Normals, and averaged
+/- 2.8% (SD: 2.3) using Method 1 and +/- 2.5% (SD: 3.8) using Method 2. Of participants with at least 1-year
follow-up data, 22% of Alzheimer's patients, 55% of MCI patients, and 67% of normals had a greater than +/-5%
cortical PIB change as shown by at least one of the two processing methods. For this subset of participants, the
estimates of the amount of 1-year PIB change between processing Method 1 and Method 2 differed by 4.7%
(SD: 2.9), although neither processing method showed a consistently greater magnitude of change. Different
image processing techniques introduce variability in longitudinal PiB estimates, likely due to differences in the
measurement of cortical regions rather than the cerebellum, but the overall directionality of these changes are