The authors have declared that no competing interests exist.
Conceived and designed the experiments: JD MLS RSJF BD. Performed the experiments: JD. Analyzed the data: JD. Contributed reagents/materials/analysis tools: JD FK KM SA. Wrote the paper: JD FK KM SA MLS RSJF BD.
¶ Please see the acknowledgments for further details of the Alzheimer's Disease Neuroimaging Initiative.
The failure of current strategies to provide an explanation for controversial findings on the pattern of pathophysiological changes in Alzheimer's Disease (AD) motivates the necessity to develop new integrative approaches based on multi-modal neuroimaging data that captures various aspects of disease pathology. Previous studies using [18F]fluorodeoxyglucose positron emission tomography (FDG-PET) and structural magnetic resonance imaging (sMRI) report controversial results about time-line, spatial extent and magnitude of glucose hypometabolism and atrophy in AD that depend on clinical and demographic characteristics of the studied populations. Here, we provide and validate at a group level a generative anatomical model of glucose hypo-metabolism and atrophy progression in AD based on FDG-PET and sMRI data of 80 patients and 79 healthy controls to describe expected age and symptom severity related changes in AD relative to a baseline provided by healthy aging. We demonstrate a high level of anatomical accuracy for both modalities yielding strongly age- and symptom-severity- dependant glucose hypometabolism in temporal, parietal and precuneal regions and a more extensive network of atrophy in hippocampal, temporal, parietal, occipital and posterior caudate regions. The model suggests greater and more consistent changes in FDG-PET compared to sMRI at earlier and the inversion of this pattern at more advanced AD stages. Our model describes, integrates and predicts characteristic patterns of AD related pathology, uncontaminated by normal age effects, derived from multi-modal data. It further provides an integrative explanation for findings suggesting a dissociation between early- and late-onset AD. The generative model offers a basis for further development of individualized biomarkers allowing accurate early diagnosis and treatment evaluation.
Establishing an accurate diagnosis of Alzheimer's disease has been a major challenge in the past decades. With an increasing amount of studies aiming at detection and validation of imaging biomarkers for this disease, many apparently controversial findings have been reported over the time. The failure of current strategies to provide a consistent explanation for these differential findings motivates the necessity to develop new integrative approaches based on multi-modal data that capture various aspects of disease pathology. In our study we propose such a generative model providing a comprehensive approach towards integration of previously published differential findings in early- and late-onset AD. We believe that our analytical strategy not only provides the link between imaging biomarkers and clinical phenotype considering the effects of aging, but could also lead to new areas of research in terms of creation of new, individualized biomarkers for a more accurate diagnosis of Alzheimer's disease.
Neuroimaging studies using [18F]fluorodeoxyglucose positron emission tomography (FDG-PET) and structural magnetic resonance imaging (sMRI) provide substantial evidence of high sensitivity for early detection and progression assessment in Alzheimer's disease (AD) at a group and single subject level
The question that needs to be answered is whether age and symptom severity indeed account for most of these discrepancies. In other words, what are the relative contributions of age and disease to the anatomical patterns of abnormality of structure and function? Investigation of these relationships may also provide clues to another long-time controversy – can the observed differences between young and aged AD patients be regarded as a continuum or is there a clear separation into two cohorts dependent on separate pathological mechanisms?
Recent studies have suggested that AD related brain changes may be similar to those associated with healthy aging. If so, this could explain age overestimation determined from sMR images from AD patients
Another controversially discussed issue is the relative capability and sensitivity of FDG-PET and sMRI to detect AD related pathology. Recent studies provided evidence for the superiority of each of the two imaging modalities as compared to the other to detect AD related pathology
To address these issues and questions we generate group level anatomical models of pathophysiological changes observed in AD using FDG-PET and sMRI data. In these models we account for PVE and integrate disease-, age- and symptom severity-associated changes in AD patients. We further dissociate them from healthy aging related changes using a combination of voxel-based general linear models (GLMs). We additionally assume that AD-induced changes are added to changes observed in healthy aging. We use the models to generate age- and symptom severity- specific whole-brain patterns of glucose hypometabolism and atrophy. To validate the obtained model and to address the questions described above, we contrapose the models' predictions in terms of anatomical plausibility to findings reported in previous studies investigating age- and AD-related changes. Thereby, we aim to provide at a group level an integrative explanation for the controversial findings described above regarding spatial characteristics and magnitude of glucose hypometabolism and atrophy in AD. We further make conclusions on the relative capability of FDG-PET and sMRI to predict AD related pathology at a group level.
We hypothesize, on the basis of the above considerations
To derive a generative model of age and symptom severity related changes, we extracted from the Alzheimer's disease Neuroimaging Initiative (ADNI) database (
Controls | AD | T-test (df,t,p) | |
Number | 79 | 80 | - |
Male/Female | 41/38 | 40/40 | - |
Age (years) | 75.8±4.9 | 75.7±7.0 | 157,0.1,.887 |
Age range (years) | 62–87 | 55–88 | - |
MMSE (score) | 28.7±1.6 | 23.6±2.2 | 157,16.6,<.001 |
Mean ± standard deviation. AD Alzheimer's disease, con converters, MMSE Mini Mental State Examination, noncon non-converters.
The ADNI was launched in 2003 by the National Institute on Aging (NIA), the National Institute of Biomedical Imaging and Bioengineering (NIBIB), the Food and Drug Administration (FDA), private pharmaceutical companies and non-profit organizations, as a $60 million, 5- year public-private partnership. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer's disease (AD). Determination of sensitive and specific markers of very early AD progression is intended to aid researchers and clinicians to develop new treatments and monitor their effectiveness, as well as lessen the time and cost of clinical trials. The Principal Investigator of this initiative is Michael W. Weiner, MD, VA Medical Center and University of California – San Francisco. ADNI is the result of efforts of many co-investigators from a broad range of academic institutions and private corporations, and subjects have been recruited from over 50 sites across the U.S. and Canada. The initial goal of ADNI was to recruit 800 adults, ages 55 to 90, to participate in the research, approximately 200 cognitively normal older individuals to be followed for 3 years, 400 people with MCI to be followed for 3 years and 200 people with early AD to be followed for 2 years.” For up-to-date information, see
The sMRI dataset included standard T1-weighted images obtained with different scanner types using a 3D MP-RAGE (magnetization-prepared 180 degrees radio-frequency pulses and rapid gradient-echo) sequence varying in TR and TE (repetition and echo time) with an in-plane resolution of 1.25×1.25 mm and 1.2 mm slice thickness acquired at 1.5T magnetic field strength. All raw data were pre-processed to correct for distortion and B1 non-uniformity as described on the ADNI webpage (
We analysed FDG-PET data for subjects who also underwent sMRI scans. FDG-PET data were acquired with different PET-scanner types according to one of three different protocols: 1) dynamic: a 30 min six-frame acquisition (6 five-minute frames), with scanning from 30 to 60 min post FDG injection; 2) static: a single-frame, 30 min acquisition with scanning 30–60 min post injection; and 3) quantitative: a 60 min dynamic protocol consisting of 33 frames, with scanning beginning at injection and continuing for 60 min. The majority of the scans in the ADNI study were acquired with the first acquisition protocol. Images further differed in resolution, orientation, voxel and image dimensions and count statistics. The frames from 30 to 60 minutes post injection were spatially realigned to minimize inter-frame motion artefacts and a mean image of these frames was calculated for each subject. These mean images were used for further analysis.
All data processing steps were carried out using the SPM5 software package (Statistical Parametric Mapping software:
All statistical analyses were also carried out using the SPM5 software package and Matlab 7.7. The effect of aging in healthy control subjects was estimated separately for FDG-PET and sMRI with voxel-wise linear regressions. To obtain the healthy aging component of our generative model we used the beta coefficients of aging in healthy controls to simulate voxel-wise changes in both imaging modalities for the age range 50 to 80 years (
a) Schematic representation of age-related changes in one voxel (in %) considering GM volume at the age of 50 years as baseline. b) Schematic representation of changes related to healthy aging (black line) and age-related differences in AD (red line) in one voxel. Intersection age (dotted line) represents the age at which healthy aging in this voxel becomes similar to changes observed in AD. The hinge in the red line (aging in AD) at the intersection point indicates that after the intersection age, according to our assumption of the additive impact of AD related processes to healthy aging, the healthy aging model would apply in AD patients as no pathological processes in terms of atrophy or glucose hypometabolism are longer observable after this time point. c) Decrease (in %) in GM volume observed in an exemplary voxel in AD depending on the constellation of age and symptom severity (MMSE) relative to the baseline provided by healthy aging (violet line). AD Alzheimer's disease, GM grey matter, MMSE Mini Mental State Examination.
To dissociate healthy aging- and AD-related changes, the variance in glucose utilization and GM atrophy explained by healthy aging was removed by voxel-wise linear regressions from all imaging data used for further models (control subjects and AD patients for both FDG-PET and sMRI;
Further, we determined at what age the changes related to normal aging become similar to those found in AD. For this purpose, a third GLM, using age as the only factor, was calculated in the AD cohort for both FDG-PET and sMRI data. The ages at which the separate regressions for normal and AD associated aging intersect (referred to as intersection ages) were calculated on a voxel-wise basis (
All three voxel-based models are based on a previously validated assumption of linearity between healthy aging- and AD symptom severity- related changes with FDG-PET and sMRI data.
Group comparisons of AD patients and control subjects for age and symptom severity were carried out using T-tests with a significance threshold of p<.05. Group differences regarding gender were evaluated using a chi-square test for independent samples. The statistical analyses were performed with SPSS 17.0 (
AD patients and control subjects did not differ in age [t(157) = 0.1;p = .887]. As expected MMSE differed significantly between the groups [t(157) = 16.6;p<.001]. The comparison of AD patients and control subjects in relation to sex showed no statistical differences [χ2(1) = 0.06;p = .811].
The generative model for healthy aging based on sMRI reveals a widespread pattern of grey matter volume reductions sparing only bilateral dorsal primary sensorimotor regions, brainstem, lateral thalamus and the dorsal part of caudate nucleus (
FDG-PET [18F]fluorodeoxyglucose positron emission tomography, GM grey matter, MRI structural magnetic resonance imaging.
The equivalent FDG-PET based model demonstrates a specific pattern of age-related metabolic changes with an age related decrease in glucose utilization of more than 10% per decade in bilateral parietal, occipital, sensorimotor, premotor, dorsolateral prefrontal and anterior insular cortices. Additionally, we see a major reduction in glucose metabolism in bilateral posterior putamina and in the left dorsal caudate nucleus (
We report a negative relationship between symptom severity and both metabolism and GM volume (
AD Alzheimer's disease, FDG-PET [18F]fluorodeoxyglucose positron emission tomography, MMSE Mini Mental State Examination.
AD Alzheimer's disease, MMSE Mini Mental State Examination, MRI structural magnetic resonance imaging.
Symptom severity related GM volume changes at age 60 years were detected throughout the brain with greatest degrees of atrophy seen in bilateral parietal, temporal, occipital, dorsolateral prefrontal, posterior cingulate and premotor cortices, the precuneus, dorsal caudate nucleus, amygdala and hippocampus. At 80 years of age, the greatest, bilateral, symptom severity related atrophy was found in parietal, temporal, occipital, primary sensorimotor and dorsolateral prefrontal cortices, the hippocampus and thalamus.
We show symptom severity- related glucose metabolism reductions bilaterally in posterior temporal, parietal, lateral occipital, dorsolateral prefrontal and premotor cortices and in the precuneus. Symptom severity related hypometabolism is less extensive at higher than lower ages.
In general, regional decreases in metabolism and grey matter volumes relative to healthy aging are significantly more pronounced in the lower compared to higher age range in AD. In the model, we observe substantial age-dependant differences in terms of hypometabolism and atrophy in AD patients compared to a healthy aging baseline even at an MMSE score of 30. At the age of 60 years these differences are bilaterally restricted to inferior frontal gyrus, premotor regions, inferior and medial temporal gyrus, cerebellum, rectal gyrus and to left parietal regions. Regions showing an initial difference in this age range do not correspond well to the anatomical pattern observed in later symptom severity stages and remain rather less affected compared to other regions. Initial differences observed at the age of 80 years are located in bilateral parietal, bilateral hippocampal and left sensorimotor regions and in both caudate nuclei.
We observe a more consistent anatomical pattern of initial differences in hypometabolism at a MMSE score of 30. For the whole age range of 60–80 years, initial glucose hypometabolism is observed in bilateral parietal, inferior temporal and posterior cingulate cortices, posterior thalamus and the precuneus. Additionally, at 80 years of age we demonstrate significant differences in bilateral primary sensorimotor and premotor regions and in the anterior temporal lobes. All regions showing initial glucose hypometabolism, except for the posterior thalamus, also show the steepest symptom severity-related metabolic decline. Regions hypometabolic only at age 80 show no specific symptom severity-related decline.
To dissociate brain changes related to healthy aging from AD pathology at specific ages we computed the intersection age of models for healthy aging and aging in AD at the voxel level (
AD Alzheimer's disease, FDG-PET [18F]fluorodeoxyglucose positron emission tomography, MRI structural magnetic resonance imaging.
With sMRI, the linear regression model describing the relationship between age, symptom severity and structural changes in AD showed a significant correlation for each factor with atrophy (
Only clusters are shown exceeding a significance threshold of p = 0.001 uncorrected on voxel level and p = 0.05 FWE-corrected on cluster level. AD Alzheimer's disease, FDG-PET [18F]fluorodeoxyglucose positron emission tomography, MRI structural magnetic resonance imaging.
With FDG-PET we found a significant positive correlation between age and glucose metabolism in bilateral temporal and parietal regions. There were no other significant correlations with age or symptom severity.
Inclusion of a quadratic relationship with age or MMSE into the models revealed significant positive correlations of glucose metabolism with a quadratic age coefficient only in left dorsal parietal cortex (
In this study we demonstrate that a generative model captures the anatomical and metabolic features associated with AD and healthy aging in the age range 60 to 80 years accurately and robustly. The model differentiates between the effects of aging and symptom severity in AD patients. It also provides a means to test for interactions between them, as exemplified here with disease progression and patient age. The age-dependant differential sensitivity of structural and metabolic scanning for the detection of AD pathology we demonstrate emphasises the ability of our generative model to infer expected age- and symptom severity- specific changes recorded with both imaging modalities.
Our study's three main findings support and extend observations made previously. Firstly, the spatial location of changes related to different ages and symptom severities in AD was mostly consistent with regions reported in the literature
Our second result is that in general, the magnitude of healthy aging related imaging changes between 50 and 80 years is comparable to that of changes associated with an increase in AD symptom severity at any age. Substantial regional overlap in both hypometabolism and atrophy was observed between a network of areas affected by healthy aging and that identified in AD. To interpret this overlap it is important to note that all changes related to age and symptom severity in AD we report here were calculated after removing the variance explained by healthy aging. That means that all age-related differences can be interpreted as an add-on required to normal age-related changes to induce a predefined symptom severity at the corresponding age. In previous research comparing differently aged AD groups with age-matched healthy control subjects a split of AD into different subgroups
A third result is in the dissociation of age- and disease- related processes inferred from the intersection ages of healthy aging and AD models. While in previous research some studies successfully applied whole-brain approaches to discriminate dementia patients from control subjects
The obtained models could be used to improve early AD detection e.g. by training automated classifiers on age- and symptom severity- specific pathological AD patterns (features) extracted by thresholding the obtained AD model. They could also be applied directly in clinical assessment by evaluating the similarity of observed pathology in any individual to age- and symptom severity- specific patterns generated using the AD model. Thereby, to enable individual assessment, percent signal difference maps could be calculated between each subject's imaging data and imaging data generated using the healthy aging model. However, both of these approaches require careful evaluation in future studies prior to clinical application.
Nonetheless, a valid interpretation of our results needs to consider the effects of several other assumptions and limitations. First, the sensitivity of the generative model for detecting and predicting AD pathology depends on the accuracy of the model of healthy aging. The age- related, widespread patterns of brain atrophy and hypometabolism we find are consistent with previous findings
A major advantage of our approach is the generalizability of our model to any constellation of age and symptom severity. A further difference to conventional approaches is that the method we propose relies on a voxel-wise group mean, ignoring the variance and so allowing detection of quite minor group differences. This type of analysis, though more flexible than conventional statistics, nevertheless requires caution in the interpretation of results. Minor differences between groups of healthy subjects and AD patients could still be due to random effects unrelated to AD. However, as we model disease progression (albeit based on cross-sectional data) we would expect the magnitude of differences to increase with higher symptom severity. This assumption suggests that local differences between AD patients and control subjects that are not so correlated with it are more likely to be due to random artefacts than disease related pathology.
For all three models used in our study, we made the assumption that healthy aging affects AD patients in the same way as healthy subjects in terms of hypometabolic and atrophic changes, with AD pathology additive to changes associated with healthy aging. This assumption is in line with current views that AD as a pathological process is unrelated to healthy aging.
A further issue of interpretation of interactions is the recognised inaccuracy of clinical diagnosis of AD, which may differ in younger
Most of these problems are also common to standard statistical methods of evaluation of group differences. In group statistics one often uses a high significance threshold to avoid false positive results thus minimising their effect on between group differentiations. By contrast, our approach also provides an opportunity to evaluate the impact of possible confounding effects, such as age in this case, on the discrimination between AD patients and control subjects.
(PDF)
We thank Jessica Peters for preparing parts of the data used in this project.
Data used in preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.ucla.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found in