The study participants comprised 24 patients (15 men and 9 women; age range, 48–90 years) enrolled in a previous multicenter study . They were recruited from an outpatient memory clinic of the National Center of Neurology and Psychiatry, Japan. The participants had a Mini-Mental State Examination score of 19.7 ± 4.6 and a global Clinical Dementia Rating of 0.8 ± 0.4. According to National Institute on Aging and the Alzheimer’s Association criteria , 10 and 14 patients were diagnosed as having possible and probable AD, respectively.
Each subject received an intravenous injection of 215 ± 33 MBq of 18F-flutemetamol (Vizamyl, Nihon Medi-Physics). All PET acquisitions were performed using a hybrid PET/CT Biograph 16 True-point scanner (Siemens Healthineers, Erlangen, Germany). After positioning, a low-dose CT scan (kVp, 130KeV; current, 40 mA; rotation time, 1.0 s; table feed per rotation, 7.2 mm; spiral pitch factor, 0.75) was acquired to be used for the attenuation correction of the PET data. Images were reconstructed using the “H10s very smooth” kernel, a 30.0-cm reconstruction field of view, and a 2.0-mm slice interval for a resulting voxel size of 0.59 × 0.59 × 2.0 mm3. A 3D-PET acquisition (list mode) was started 61.2 ± 0.8 min after the injection of the tracer and lasted for 20 min. Image reconstruction was performed using a 3D ordered subsets expectation maximization algorithm with the following parameters: image matrix, 168; field of view, 300 mm; subsets, 21; iterations, 4; post-filter (Gaussian), 4-mm FWHM; attenuation correction, CT-based. The resulting voxel size was 2.02 × 2.06 × 2.03 mm3.
The MRI for all patients was performed on an Achieva 3.0-T MR scanner (Philips Medical Systems, Best, The Netherlands) equipped with a 32-channel coil within 42 ± 21 days before the amyloid PET. A volumetric turbo field echo T1-weighted structural sequence (300 sagittal slices; TR, 7.0 ms; TE, 3.4 ms; field of view, 260 × 240 mm; voxel size, 0.7 × 0.7 × 0.6 mm3; flip angle, 10°) was acquired for each subject.
Figure 1 shows the processing pipeline applied to quantitative analysis using the SUVR and the 100-point CL scale . This MRI-based pipeline has already been validated  using a GAAIN dataset of 11C-PiB PET images for 34 young control individuals and 45 typical AD patients downloaded from the GAAIN website. In the present study, low-dose CT instead of MRI was also used for anatomic standardization. The CL scale assigns an average value of 0 to high-certainty amyloid-negative subjects and an average of 100 to typical AD patients. First, in this pipeline, the subject MRI or CT was manually oriented and coregistered to the MNI template (avg152T1.nii) provided with Statistical Parametric Mapping (SPM) 12 software (https://www.fil.ion.ucl.ac.uk/spm). The subject PET was then manually oriented and coregistered to the coregistered subject MRI or CT. Then, the coregistered subject MRI or CT was warped into MNI space using unified segmentation in SPM12. The parameters of the deformation field in this warping were applied to the coregistered subject PET for anatomic standardization into MNI space. Using the standard VOI in GAAIN, the SUVR was calculated from 18F-flutemetamol PET counts in the global cortical target area (GAAIN, CTX VOI) and in the whole cerebellum (GAAIN, WhlCbl VOI) as the reference area. Then, a direct conversion equation (CL = 121.42 × SUVR − 121.16) was applied to convert the SUVR to the CL value, as described previously . We respectively defined the SUVRMRI and SUVRCT and CLMRI and CLCT as the use of MRI and CT for anatomic standardization.
The endpoint of this study was the measurement accuracy of the SUVRCT and CLCT when the SUVRMRI and CLMRI were respectively regarded as the gold standard.
Concordances between the SUVRMRI and SUVRCT and between the CLMRI and CLCT were assessed using Bland–Altman plots and Pearson correlation estimates. In the Bland–Altman plot of the SUVR and CL, we performed a Spearman correlation to test whether there were associations between the difference in SUVRMRI versus SUVRCT and the SUVR load and between the difference in CLMRI versus CLCT and the CL load. The SUVR and CL and their standard deviations were computed with mean absolute differences and limits of agreement. These statistical tests were performed using JMP ver. 16 (SAS Institute). In addition, to investigate regional differences in the CT-based and MRI-based standardized amyloid PET images, a paired t-test was applied to these images on a voxel basis after smoothing with an 8mm FWHM Gaussian kernel using SPM12. Results were considered significant at p < 0.001 with an extent threshold of 300 voxels without multiple comparisons.