We retrospectively selected two groups of subjects, a group of patients with AD and a group of healthy control subjects (HC). Both groups had undergone a brain 18F-FDG PET scan on a digital PET/CT (Vereos, Philips®), at the CHRU of Nancy, France, between December 2017 and September 2019.
The AD patients selected fulfilled the NIA-AA 2018 criteria for AD (14). They exhibited positive cerebro-spinal fluid biomarkers with increased phosphorylated Tau protein and reduced beta-amyloid peptide levels measured in the same laboratory (Department of Biochemistry, Molecular Biology and Nutrition, CHRU Nancy, France) using standard cut-offs (15). AD patients also underwent a routine neurocognitive assessment in the “memory clinic” of the university hospital of Nancy (France).
Healthy control subjects were age and sex matched with patients in the AD group and were also selected retrospectively. All healthy control subjects had undergone a brain 18F-FDG PET scan for cognitive assessment, but had returned a normal scan by careful visual analysis (EM, AV) and a neuropsychological assessment which was not consistent with a neurodegenerative disorder: i) normal neuropsychological tests, i.e. MMSE ≥ 27, FAB ≥ 15 and no major depressive disorders and ii) a clinical follow-up, of longer than 1 year, which showed a stabilisation and/or improvement of cognitive symptoms.
Our AD and HC groups were further compared to each other and to two control databases (a digital and a conventional database) derived from prospective studies. Individuals from these two control databases had undergone a brain 18F-FDG PET/CT performed with a conventional camera between October 2009 to May 2012 (n=19, Biograph 6, Siemens®, NCT02858167) or performed with a digital camera between December 2017 to June 2019 (n=20, Vereos, Philips®, NCT03345290) and were age and sex matched with our AD and HC groups. A flowchart summarising the constitution of the different control groups is shown in Figure 1.
Informed consent was obtained for each participant included in the selected groups. This study was approved on January 16, 2020 by the local ethics committee (NCT04163276, Study ID Numbers: 2019PI238).
Brain 18F-FDG PET
The brain 18F-FDG PET scan was recorded over a 10 (for conventional camera) to 15 minutes (for digital camera) one bed acquisition, 45 to 50 minutes after injection of 4.5 MBq/kg (conventional camera) or 2-3 MBq/kg (digital camera) of 18F-FDG. All subjects had fasted at least 6 hours prior to receiving the injection and had blood glucose levels < 10 mmol/L. All PET images were reconstructed with iterative OSEM methods, as performed in routine clinical practice, and corrected for scatter, random and attenuation with a CT scan. Reconstructed parameters included 4 iterations and 8 subsets, subsequently displayed in a 168 x 168 matrix with 2.7 x 2.7 x 2.7 mm3 voxels for the conventional PET camera (7), and 3 iterations and 15 subsets, subsequently displayed in a 256 x 256 matrix with 1 x 1 x 1 mm3 voxels for the digital PET camera (2).
Statistical Parametric Mapping
The 18F-FDG PET brain images were pre-processed using SPM12 (Wellcome Department of Cognitive Neurology, Institute of Neurology, London, UK) running on Matlab 2018a (MathWorks Inc., Sherborn, MA). After an initial step of approximate manual re-orientation and positioning to MNI space, the spatial normalisation of each PET image into the MNI space was performed by spatial normalisation of the CT scan for each subject provided by the correction of attenuation, using the method and the template of the Clinical Toolbox for SPM (https://www.nitrc.org/projects/clinicaltbx/). The voxel sizes of the written CT images were set to 1 x 1 x 1 mm3 for the digital controls and to 2 x 2 x 2 mm3 for the conventional controls. Each CT spatial normalisation procedure was subsequently applied to the respective PET images. Voxels of PET images recorded with the conventional camera during this step were therefore resampled from 2.7 x 2.7 x 2.7 mm3 to 2 x 2 x 2 mm3 whereas voxel sizes of the digital databases were not modified (1 x 1 x 1 mm3). To enable voxel-to-voxel analysis with these two control databases, AD and HC group datasets were normalised using the two voxel sizes. Partial volume effect corrections were applied to PET images using the Müller-Gärtner (MG) method provided by the PETPVE12 toolbox (16). White and grey matter segmentations needed for the 3- compartmental voxel-wise MG method were realised on CT scans using SPM segmentation tools, after careful visual analysis at the individual level to check the accuracy of this segmentation. The cerebellum was used as reference for the intensity normalisation of PET images because normalisations other than the proportional scaling have been proposed (17) and because the cerebellum is associated with a more accurate discrimination of patients with AD compared to controls (18). All regions of interest (ROIs) used for intensity normalisation (all the cerebellum and vermis ROIs for the cerebellum, and all the pre- and post-central brain areas for the sensorimotor cortex) were extracted from the AAL atlas (19) after spatial normalisation to limit the inter-individual anatomical heterogeneity. Finally, PET images were smoothed with an isotropic 3D Gaussian kernel of 12 mm FWHM to blur individual variations in gyral anatomy. Visual inspections of the images at the different stages of the pre-processing procedure ensured the quality and convergence of the different methods applied.
Semi-Quantitative Analyses (SQA) were performed at the group and individual level on a voxel-by-voxel basis using two-sample t-tests with an inclusive AD mask (20). At the group level, AD and HC groups were compared with the conventional and digital controls using age and sex as covariates (clusters of decreased metabolic activity observed at p<0.001 for the voxel, cluster volume corrected by using the expected volume provided by SPM and based on the random field theory). We used exclusive masks to compare results obtained with SQA to conventional or digital controls. For AD, an exclusive mask corresponding to the SPM-T map results of SQA to conventional controls was applied to the SQA of digital controls to highlight the additional clusters visualised with the digital system compared to the conventional system (and vice versa for the HC population).
At the individual level, each subject in the AD and HC group, was individually compared to the digital and conventional controls using a fully automated analysis as well as visual analyses (clusters of decreased metabolic activity observed at p<0.005 for the voxel, cluster volumes corrected to 800 mm3 (6) (21)). All clusters identified with SPM at the individual level were considered significant for the fully automated analysis.
The precise identification of each structure located by its MNI coordinates, its respective volume, and T-max intensity were extracted by using the report provided by the SPM xjView toolbox (http://www.alivelearn.net/xjview).
Visual ratings at the individual level:
For the visual analyses, the SPM T-maps were projected onto three-dimensional rendering of T1-weighted MRI images using SPM surface rendering tool and onto 12 two-dimensional slices of T1-weighted MRI images using the Slice Display toolbox (22) (axial orientation, inter-slice spacing of 1 cm). Representations were reviewed by three experienced observers (EM, EG and AV), who were blinded to the patient’s clinical data. Raters were forced to give a dichotomous reading: Alzheimer’s disease diagnosis or not pathological. A pattern of diffuse hypometabolic areas within the areas known to be involved in AD (mainly the bilateral posterior associative areas) was considered a positive scan. At the individual level, results were expressed as a consensual analysis for the positive diagnosis of AD.
Categorical variables are expressed as percentages and continuous variables as means and standard deviations. Due to the non-normality of variable distributions, Chi-2 and Kruskal-Wallis tests were performed for comparisons of categorical and continuous variables, respectively. For the comparisons of diagnostic performances at the individual level, Mc Nemar tests were used with corrections for multiple comparisons. A p-value <0.05 was considered as significant. All tests were performed with SPSS (SPSS Statistics for Windows, Version 20.0. Armonk, NY: IBM Corp).