Processed FDG-PET data from: A computational model of neurodegeneration in Alzheimer’s disease
Cite this dataset
Jones, David (2022). Processed FDG-PET data from: A computational model of neurodegeneration in Alzheimer’s disease [Dataset]. Dryad. https://doi.org/10.5061/dryad.msbcc2g0n
Disruption of mental functions in Alzheimer’s disease (AD) and related disorders is accompanied by selective degeneration of brain regions. These regions comprise large-scale ensembles of cells organized into systems for mental functioning, however the relationship between clinical symptoms of dementia, patterns of neurodegeneration, and functional systems is not clear. We developed a model of the association between dementia symptoms and degenerative brain anatomy using F18-fluorodeoxyglucose (FDG) PET and dimensionality reduction techniques patients with AD. This data and code package contains preprocessed FDG-PET images from 423 subjects across the Alzheimer’s disease spectrum and the MATLAB code to produce eigenbrains from this data.
All participants or their designee provided written consent with approval of the Mayo Clinic Foundation and Olmsted Medical Center Institutional Review boards. All participants in the Mayo Clinic Rochester Alzheimer’s Disease Research Center (ADRC) and the Mayo Clinic Study of Aging (MCSA) that met our inclusion criteria were included in this study. The Mayo Clinic Rochester ADRC is a longitudinal cohort study that enrolls subjects from the clinical practice at Mayo Clinic in Rochester, MN.
1 The MCSA is a population-based study of cognitive aging among Olmsted County, MN residents
2. Enrolled participants are adjudicated to be clinically normal or cognitively impaired by a consensus panel consisting of study coordinators, neuropsychologists, and behavioral neurologists. Methods for defining clinically unimpaired, mild cognitive impairment and dementia in both studies conform to standards in the field 3-5. MCSA study participants receive renumeration of USD 100 as part of study participation. Both the MCSA and the ADRC studies offer assistance with ground transportation cost associated with study participation and USD 50 for participation in PET scanning portions of the study.
Inclusion criteria for this study consisted of 1) a CDR global score greater than zero, 2) presence of amyloid plaques, defined as amyloid-PET standard uptake value ratio (SUVR) greater than 1.5, and 3) had high quality MRI, amyloid-PET, and FDG-PET data available for analysis. A higher more conservative SUVR cut point was used for defining amyloid-PET positivity to avoid false positives 1.
Structural Magnetic Resonance Imaging:
MRI was performed on one of three compatible 3T systems from the same vendor (General Electric, Waukesha, WI, USA).1 A 3D magnetization prepared rapid acquisition gradient echo (MPRAGE) structural imaging sequence developed for the Alzheimer's Disease Neuroimaging Initiative (ADNI) study was acquired 6. All images were acquired using an 8-channel phased array head coil. Post-processing to correct for gradient distortion correction and processing has been validated in multiple studies, shown to give consistent stable results in ADNI data, and geometric fidelity after correction is independent of scanner 7,8. Parameters were: TR/TE/T1, 2300/3/900 msec; flip angle 8°, 26 cm field of view (FOV); 256 × 256 in-plane matrix with a phase FOV of .94, and slice thickness of 1.2 mm. These MPRAGE parameters have been held invariant since approximately 2008. This structural MRI was used for preprocessing PET data. MRI data is not included with this dataset.
PET Acquisition and Preprocessing:
The amyloid-PET imaging was performed with C-11 Pittsburgh Compound B 9 and FDG-PET with F-18 fluorodeoxyglucose. PET images were acquired using 1 of 2 PET/CT scanners (DRX; GE Healthcare). A computed tomography scan was obtained for attenuation correction. These images were usually acquired on the same day with 1 hour between amyloid-PET and FDG-PET acquisitions. Subjects were prepared for FDG-PET in a dimly lit room, with minimal auditory stimulation. Amyloid-PET images consisted of four 5-min dynamic frames from 40 to 60 min after injection. FDG-PET consisted of four 2-min dynamic frames acquired from 30 to 38 min after injection. PET sinograms were iteratively reconstructed into a 256 mm FOV. The pixel size was 1.0 mm and the slice thickness 3.3 mm. Standard corrections were applied.
The global amyloid-PET SUVRs were calculated as previously described 10. The amyloid-PET images are not incuded with this dataset. The FDG-PET image volumes of each subject were coregistered to the subject’s own T1-weighted MRI scan, using a 6 degree-of-freedom affine registration with mutual information cost function. Each MRI scan was then spatially normalized to an older adult template space 11 using a unified segmentation and normalization algorithm 12 with transforms applied to co-registered FDG-PET images. These spatially normalized images were then intensity normalized to the pons and spatially smoothed with a 6 mm full-width half-maximum Gaussian kernel.
Between-subject variability projection and reduction:
The unsupervised machine learning framework, Between-subject variability Projection and Reduction (BPR), was designed to capture pathophysiologic information present in between-subject variability in a disease parameter of interest. The singular value decomposition (SVD) at the heart of the data reduction portion of the algorithm is widely used and interpretable, but other methods could be used depending on the framing of the problem at hand. The goals of this framework also motivate data preprocessing decisions that focus on between-subject variance within the class being studied rather than variance in the observed modality under investigation or variance relative to classes not being studied. This algorithm conceptualizes multivariate medical data from an individual as representing a particular parameterization of a (patho)physiological process of interest and uses within-class individual differences in this parametrization to define a high dimensional parameter space that contains a smaller dimensional subspace manifold that describes common features of the disease generating processes of interest. This lower dimensional subspace can be isolated in many ways, but ideally the dimensionality reduction technique used would retain interpretability in order to promote understanding of the pathophysiology of interest and be able to meaningfully place new subjects into the learned subspace and make interpretable predictions about clinical variables of interest.
In the present study, we assume that macroscale glucose uptake patterns in cognitively impaired individuals with amyloid plaque deposits represent a parameterization of macroscale AD pathophysiology. We then isolated the between-subject variability of interest to this study from these preprocessed FDG-PET scans in the following way. The preprocessed FDG-PET images are three-dimensional arrays of voxel intensities that correspond to SUVR values in a standard template space. Taking only the voxel intensities that fall within the set of voxels that have a greater than 15% probability of being gray matter in template space, this three-dimensional array can be reduced to a one-dimensional vector. To isolate subject effects, each element is non-parametrically standardized by the median and interquartile range for that element across subjects. Subject-wise centering of each image is then performed. This can then be used to represent the individual differences of interest in the brain images between each image pair, or between subject variance, by calculating the subject-wise covariance matrix.
This high-dimensional projection of individual differences can be represented as an eigendecomposition, using the singular-value decomposition, such that the eigenvectors of subject-wise covariance determine the linear combination of the M set of FDG-PET images that produce image space eigenvectors, or eigenbrains given that they can be ordered into a three-dimensional configuration corresponding to the original brain images, as previously described for the eigenfaces facial recognition algorithm for two-dimensional facial recognition 13. This algorithm demonstrates how individual differences in multivariate patterns in brain images can be mapped back into the original image space in the form of a compact lower-dimensional basis-set of eigenbrains (EBs). This allows for a highly interpretable understanding of the parameterization of a disease process affecting the individuals included in the analysis. The first 10 EBs explained 51% of the variance in the dataset.
1) Jones, D. T. et al. Tau, amyloid, and cascading network failure across the Alzheimer's disease spectrum. Cortex; a journal devoted to the study of the nervous system and behavior 97, 143-159, doi:10.1016/j.cortex.2017.09.018 (2017).
2) Roberts, R. O. et al. The Mayo Clinic Study of Aging: design and sampling, participation, baseline measures and sample characteristics. Neuroepidemiology 30, 58-69, doi:10.1159/000115751 (2008).
3) Albert, M. S. et al. The diagnosis of mild cognitive impairment due to Alzheimer's disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 7, 270-279, doi:10.1016/j.jalz.2011.03.008 (2011).
4) McKhann, G. M. et al. The diagnosis of dementia due to Alzheimer's disease: recommendations from the National Institute on Aging-Alzheimer's Association workgroups on diagnostic guidelines for Alzheimer's disease. Alzheimers Dement 7, 263-269, doi:10.1016/j.jalz.2011.03.005 (2011).
5) Petersen, R. C. Mild cognitive impairment as a diagnostic entity. J Intern Med 256, 183-194, doi:DOI 10.1111/j.1365-2796.2004.01388.x (2004).
6) Jack, C. R., Jr. et al. The Alzheimer's Disease Neuroimaging Initiative (ADNI): MRI methods. J Magn Reson Imaging 27, 685-691, doi:10.1002/jmri.21049 (2008).
7) Gunter, J. L. et al. Measurement of MRI scanner performance with the ADNI phantom. Med Phys 36, 2193-2205, doi:10.1118/1.3116776 (2009).
8) Vemuri, P. et al. Accelerated vs. unaccelerated serial MRI based TBM-SyN measurements for clinical trials in Alzheimer's disease. Neuroimage 113, 61-69, doi:10.1016/j.neuroimage.2015.03.026 (2015).
9) Klunk, W. E. et al. Imaging brain amyloid in Alzheimer's disease with Pittsburgh Compound-B. Ann Neurol 55, 306-319, doi:10.1002/ana.20009 (2004).
10) Lowe, V. J. et al. Comparison of 18F-FDG and PiB PET in cognitive impairment. J Nucl Med 50, 878-886, doi:10.2967/jnumed.108.058529 (2009).
11) Vemuri, P. et al. Antemortem MRI based STructural Abnormality iNDex (STAND)-scores correlate with postmortem Braak neurofibrillary tangle stage. Neuroimage 42, 559-567, doi:10.1016/j.neuroimage.2008.05.012 (2008).
12) Ashburner, J. & Friston, K. J. Unified segmentation. Neuroimage 26, 839-851, doi:10.1016/j.neuroimage.2005.02.018 (2005).
13) Turk, M. & Pentland, A. Eigenfaces for recognition. J Cogn Neurosci 3, 71-86, doi:10.1162/jocn.1918.104.22.168 (1991).
National Cancer Institute, Award: P30 AG62677