Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (http://adni.loni.ucla.edu
). 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 MRI, PET, other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of MCI and early 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 US.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 http://www.adni-info.org
ADNI is a landmark multi-site study of biomarkers for MCI and AD that emphasizes MRI as a key modality [22
]. A hallmark of this study is the availability of imaging and clinical data to the scientific community through an easily accessible set of web sites. Criteria for participation in ADNI as a normal control include age > 55 years, no clinical or imaging indicators of neurologic or psychiatric disease, absence of a memory complaint, normal basic and instrumental activities of daily living, and normal scores on the Mini-Mental Status Examination (MMSE; [23
]), clinical dementia rating scale (CDR; [24
]), and revised Wechsler memory scale [25
]. Further protocol details are available on the ADNI web site: http://www.adni-info.org
ADNI used a standardized 1.5-T MR imaging protocol across sites that included two T1-weighted magnetization-prepared rapid gradient-echo (MP-RAGE; [26
]) imaging acquisitions, with TR 9 ms, TE 4 ms, 8° flip angle, 2562
in-plane samples, and 192 sagittal slab partitions, with a spatial resolution of 0.94 × 0.94 × 1.2 mm. For full protocol see the LONI web site: http://www.loni.ucla.edu/ADNI/Research
. Image data from 232 subjects labeled as normal were downloaded from the LONI web site on October 17, 2009. These images were B1-field corrected using the N3 algorithm [27
], corrected for gradient distortion based on a standardized phantom scanned at each site, and scaled according to the ADNI protocol before being made available by the ADNI Mayo Clinic Imaging Core to LONI for download. Three of the image data sets were discarded because of poor quality, leaving scans from 229 ADNI normal control subjects for potential analysis.
Clinical data on these subjects were obtained from the ADCS web site on June 3, 2011, the cutoff date for follow-up in the current study. Criteria for inclusion in the present study were: 1) diagnosis of “Normal” at baseline, global Clinical Dementia Rating (CDR) score of 0, and CDR box score of 0 for both memory and orientation at the screening visit; 2) assessment at six months that included a CDR; and 3) follow-up for at least one year. These criteria were chosen because the goals of the study were to examine well-characterized normal subjects who were followed longitudinally with a minimum of at least two visits after the screening and baseline visits. Exclusions based on the above criteria (total n = 21) were made for the following reasons: follow-up less than a year (n=12), no 6-month evaluation (n=8), and CDR box score 0.5 for Orientation alone at month 6 visit (n=1).
Using both the baseline and follow-up data, comparison groups were defined as follows: 1) Normal Nonconverters (NN, n = 155) - Clinical diagnosis of “Normal”, and both global CDR of 0 and CDR box scores for memory and orientation of 0 at baseline and at all follow-up visits; and 2) Memory Complainer (MC, n = 53) - Normal at baseline, but CDR box score for memory of 0.5 or greater at any time after the initial (screening) visit. MC is not equivalent to a diagnosis of MCI; 37 of these 53 subjects remained classified as clinically normal during follow-up (subgroup MC_N), and 16 were “converters” from normal at baseline to MCI or AD as diagnosed by the clinicians in ADNI (subgroup MC_C). Median time from baseline to conversion in MC_C subjects was 36 months (IQR 24–36 months). lists the demographic characteristics of the NN and MC subjects. The median follow-up time was four years in both groups. presents key results of the clinical and cognitive assessments in ADNI in all subjects at baseline and in the remaining subjects seen at the 48-month visit.
Table 1 Baseline characteristics of study subjects. MC subgroups: MC_N = MC subjects who remained normal and did not convert to MCI during the observation period; MC_C = MC subjects classified as normal at baseline but who converted to MCI during longitudinal (more ...)
ADNI Assessment Results. The median follow-up, 48 months, was chosen for longitudinal comparison to preserve statistical power lost to subject attrition over time.
A pre-defined brain region mask for validation was developed in a prior study using images from an independent group of 136 normal aged subjects who did not participate in ADNI (Full details provided in [17
]). All 136 subjects were normal at baseline; of these 113 remained normal throughout the period of longitudinal observation (5.4 years), and 23 subsequently developed cognitive impairment an average 4.0 years after baseline (MCI or AD). The brain region mask represents the region of significant volume difference between subjects who remained normal and those who converted to MCI or AD (Region illustrated on a single subject T1 axial image in ). This region (called AMTR for short) mainly comprised the anteromedial temporal lobes bilaterally, with an additional small region within the left angular gyrus, and even smaller clusters of voxels in the anterolateral temporal lobes. The techniques for image analysis included optimized voxel-based morphometry using multi-channel MRI images for segmentation and custom templates for registration and analysis in SPM2 (Wellcome Trust Centre for Neuroimaging; http://www.fil.ion.ucl.ac.uk/spm/
). The main question addressed in the current study was whether this region could be cross-validated, by demonstrating reduced volumes in an independent longitudinal sample of baseline-normal ADNI subjects who would later develop cognitive impairment.
Figure 1 A) AMTR Region. Illustration of AMTR showing ICBM single subject T1 axial image without (left) and with (right) overlay of AMTR for this slice level. The AMTR comprises mainly amygdala and hippocampus. B) Processing steps for image normalization. Grey (more ...)
Processing of ADNI Images
All 208 ADNI images were acquired at 1.5T; no 3T images were used. Images were segmented into six partitions (grey matter, white matter, cerebrospinal fluid, and three sets of non-brain tissues) using SPM8 prior probability maps (New Segment, unified segmentation; [28
]). Segmentations were reviewed individually for quality and three images were deemed initially unsatisfactory; they were re-segmented after partially removing extracranial tissues using the Brain Extraction Tool (BET) in FSL v. 4.1.1 [31
] and then included with the remaining images. A common template was created using an SPM8 iterative diffeomorphic registration algorithm on the segmented grey matter images. The common template was registered to MNI space. Modulated grey matter mages were normalized to the registered common template using the registration flow fields and then smoothed with an 8 mm kernel. Modulated images retain volume information from the original source images. An illustration of a representative segmented image registered to this template is shown in .
Volume within AMTR for each subject was extracted using the MNI space registered mask on each smoothed, registered, modulated grey matter image. This process is illustrated in . A similar process of region volume extraction was performed using anatomic regions defined in the LONI MNI/ICBM atlas template (http://www.loni.ucla.edu/ICBM
; ). The ICBM regions chosen were amygdala, entorhinal area, hippocampus, posterior cingulate, precuneus, superior and inferior parietal lobe, superior, middle and inferior temporal gyri and fusiform gyrus. Total intracranial volume (TIV) for each subject was calculated by summing the volumes from grey matter, white matter and cerebrospinal fluid segmented images with probability threshold set at >0.49 for each image.
Figure 2 A) Extraction of volumes from subject images. The AMTR mask region (image 1) was used to extract voxel values representing grey matter volume from each subject image (images 2 & 3) by isolating only the voxels underlying the mask (image 4). Total (more ...)
A standard least-squares model was developed in JMP 9.01 (SAS Institute, Cary, NC). The model examined the effect of group and of potential influential covariates on structure volumes separately. Extracted grey matter volume within AMTR was the dependent variable with the main effect of interest being group membership (NN, MC_N, MC_C), with additional covariates age at scan, gender, education, APOE allele status (ε4 allele present or absent), and TIV as independent variables (). The same grouping variable and covariates were used in similar models with grey matter volumes extracted using standard template anatomic regions from the ICBM atlas as dependent variables ().
Regional grey matter volumes in ADNI subjects who were normal at study entry.
Receiver operator characteristic curves generated from a logistic regression model were then constructed contrasting MC_N versus NN and MC_C versus NN, with age at scan, gender, education, APOE allele status (ε4 allele present or absent), and AMTR volume as the regressors ().
Volume within AMTR is predictive of a later classification of MCI in baseline normal subjects in ADNI with 84% overall accuracy (Later MCI; MC_C). Accuracy was 71% for the entire MC group combined.
For analysis of demographic and clinical variables, one-way ANOVA with F-ratio test for significance was used for continuous variables, Pearson test on Chi Square for categorical variables, and Wilcoxon Rank Sums for nonparametric tests of potentially skewed variables. A p-value of less than 0.05 was considered significant.