We recruited 47 CN older adults from the Brain Mapping by Integrated Neurotechnologies for Disease Studies (Brain/MINDS) project (grant number 18dm0207017h0005). All individuals underwent structural magnetic resonance imaging (MRI) and 11C-Pittsburgh compound-B (PiB) and 18F-THK5351 positron emission tomography (PET) scans, as well as cognitive testing that included the Mini-Mental State Examination (MMSE), global Clinical Dementia Rating Scale (CDR), and Wechsler Memory Scale-Revised Logical Memory II (WMSR LM-II). The inclusion criteria were as follows: visually negative 11C-PiB PET results, a global CDR score of 0, an MMSE score of ≥ 26, performance within education-adjusted norms for the WMSR LM-II, no neurological or psychiatric disorders, and no medications that affect cognition. Amyloid negativity was visually assessed by a board-certified nuclear medicine specialist from 11C-PiB PET.
All participants underwent structural MRI scans on a Siemens 3-T scanner (Verio; Siemens, Erlangen, Germany) to obtain 3D sagittal T1-weighted magnetization-prepared rapid acquisition with gradient echo (MPRAGE) images (repetition time/echo time, 1.900/2.52 ms; 1.0-mm effective slice thickness with no gap; 300 slices; matrix, 256 × 256; field of view, 25 × 25 cm; acquisition time; 4 min 18 s).
All PET/computed tomography (CT) scans were acquired using a Siemens/Biograph TruePoint16 Scanner (3D acquisition mode; 81 image planes; 16.2-cm axial field of view; 4.2-mm transaxial resolution; 4.7-mm axial resolution; 2-mm slice interval). Low-dose CT scans for attenuation correction were performed prior to the PET scans. 11C-PiB PET scans were acquired as dynamic scans in LIST mode 50-70 min after a bolus injection of 555 ± 185 MBq of 11C-PiB. 18F-THK5351 scans were acquired as dynamic scans in LIST mode 40–60 min after a bolus injection of 185 ± 37 MBq of 18F-THK5351. PET/CT data were reconstructed using an iterative 3D ordered subset expectation maximization reconstruction algorithm. All MRI and PET data were acquired in the same manner as in previous studies [1, 6].
MRI and PET processing
GM images were segmented from 3D T1-weighted images using Statistical Parametric Mapping Software version 12 (SPM12; Functional Imaging Laboratory, University College London, London, UK) implemented in MATLAB 7.12. Partial volume-corrected 11C-PiB and 18F-THK5351 PET images obtained using PETPVE12 toolbox  were normalized using SPM12. Each participant’s PET images were coregistered to the corresponding T1-weighted images and normalized to the Montreal Neurological Institute (MNI) space with the Diffeomorphic Anatomical Registration Through Exponentiated Lie (DARTEL) method . After spatial normalization, the standardized uptake value ratio (SUVR) for 11C-PiB and 18F-THK5351 PET images were calculated using the individual’s positive mean uptake value of cerebellar GM as the reference region. Finally, each SUVR of PET image was smoothed using an 8-mm full width at half maximum (FWHM) Gaussian kernel. MRI and PET data were processed in the same manner as in previous studies [1, 6].
Single-subject GM networks
We resliced all native segmented GM images into 2 × 2 × 2-mm isovoxels to standardize voxel sizes and reduce dimensionality. Single-subject GM networks were extracted based on intracortical similarity from native space GM segmentations using a previously described fully automated method (https://github.com/bettytijms/Single_Subject_Grey_Matter_Networks; version 20150902) . Nodes were defined as small regions of interest in the brain (3 × 3 × 3 voxel cubes, corresponding to 6 × 6 × 6 mm3). Connectivity was defined by high statistical similarities quantified with Pearson’s correlations across the GM density values of corresponding voxels between any two nodes. Then, each node was rotated by a θ angle at multiples of 45° and reflected over all axes to identify the maximal similarity value with the target node. To construct unweighted and undirected graphs, the GM similarity matrices were binarized after determining a threshold. The threshold for each graph was determined using false discovery rate technique to correct for multiple comparisons which is based on the random permutation method to ensure similar chance to include average 5% spurious correlations for all subjects . The following four local network measures were calculated: betweenness centrality (i.e., the proportion of the shortest paths that run through a node), clustering coefficient (i.e., the level of interconnectedness of neighboring nodes), characteristic path length (i.e., the shortest distance between two nodes), and degree (i.e., the number of edges per node).
To compare each participant’s local network measures at the cubic level, we superimposed the images on the corresponding resliced GM of the MNI space. For single-subject GM networks, each participant’s local network measure images were smoothed using a 10-mm FWHM Gaussian kernel in the same manner as in the previous study . This value (10 mm) was determined by nearly doubling the resolution of one side of a cube (6 mm).
Voxel-wise correlations between 18F-THK5351 PET and GM network
To evaluate relationships between 18F-THK5351 and local network measures, we used the Biological Parametric Mapping (BPM) toolbox . The toolbox allows voxel-wise correlations across two imaging modalities based on the general linear model. We analyzed the correlations between 18F-THK5351 and four local network measures (betweenness centrality, clustering coefficient, characteristic path length, and degree). The results with the following criteria were deemed significant: a height threshold of p < .05 (family-wise error [FWE] corrected) with an extent threshold of 100 voxels.