Voxel-wise detection of amyloid deposition with Pittsburgh compound B (PIB) in normal aging: A partial volume correction-based approach

Background and Aims: Identification of voxel-wise amyloid deposition in normal aging may be hampered by high inter-subject variability and the apparent greater retention of PIB in white matter than gray matter. We employed a new approach to compare observed binding potential (BP) maps to partial volume correction (PVC)-predicted BP maps that served as individuals' own control, to cope with inter-subject variability while accounting for gray-white matter composition at the voxel level. Methods: Twenty-eight nondemented participants of Baltimore Longitudinal Study of Aging (age: 79.2 ± 8.0; 11 males and 17 females) underwent one 90-min dynamic PIB-PET scan. BP maps of PIB scans were constructed by the bolus-plus-infusion transformation (BPIT; Kuwabara et al., 2002). Gray matter, white matter, and cerebro-spinal fluid space segments were identified on MRI, transferred to PET space, and smoothed by PET camera specific Gaussian kernels to reproduce the partial volume effect. The resulting geometric transformation matrix (M; Rousset et al., 1998), is an n by 3 matrix where n is the number of voxels covering the three segments. PV-corrected BP values of the three segments (T, a 3 by 1 matrix) were obtained by left matrix division using Gaussian elimination. The PVC-predicted BP map was constructed by inserting the M*T product to the PVC voxels and -1 to the remaining voxels. Thus, PVC-predicted BP maps represent weighted mean BP values of the three segments at the voxel level assuming a homogenous BP value for each segment throughout the brain. Comparison of observed BP maps to PVC-predicted BP maps, using SPM2, allows for exploring regions that show significant deviation from the homogeneity assumption. Results: PV-corrected BP values were 0.14 ± 0.15 (unit-less; mean ± SD) for gray matter, 0.54 ± 0.10 for white matter, and -0.73 ± 0.12 for cerebro-spinal space. Table lists four clusters (observed BP values > PVC-predicted BP values) identified with SPM (p<0.05, corrected; volume > 1 ml) across the 28 subjects. Conclusions: The approach successfully identified four regions of relative increases in PIB binding in nondemented older adults with stringent statistical criteria. The results suggest the PVC-based approach can detect subtle increases in PIB binding in nondemented subjects with a low cortical BP value despite a several fold higher white matter BP value. Further studies on demented populations will be required to understand the clinical significance of PIB accumulation at this early stage of amyloid deposition.