Predicting Future Cognitive Decline using Tensor-Based Morphometry

By Megan Griffin,2014-11-25 09:30
9 views 0
Predicting Future Cognitive Decline using Tensor-Based Morphometry

    Tensor-Based Morphometry as a Neuroimaging Biomarker for Alzheimer’s Disease:

    An MRI Study of 676 AD, MCI, and Normal Subjects

     1111Xue Hua, Alex D. Leow MD PhD, Neelroop Parikshak, Suh Lee, Ming-Chang Chiang MD 1123,4,5PhD, Arthur W. Toga PhD, Clifford R. Jack Jr MD, Michael W. Weiner MD 1 Paul M. Thompson PhD

    and the Alzheimer's Disease Neuroimaging Initiative*

     1Laboratory of Neuro Imaging, Dept. of Neurology, UCLA School of Medicine, Los

    Angeles, CA 2Mayo Clinic College of Medicine, Rochester, MN 34Dept. Radiology, and Dept. Medicine and Psychiatry, UC San Francisco, San Francisco,

    CA 5Department of Veterans Affairs Medical Center, San Francisco

    Manuscript submitted to NeuroImage: April 6, 2008

    Please address correspondence to:

    Dr. Paul Thompson, Professor of Neurology

    Laboratory of Neuro Imaging, Dept. of Neurology,

    UCLA School of Medicine

    Neuroscience Research Building 225E

    635 Charles Young Drive, Los Angeles, CA 90095-1769, USA

    Phone: (310) 206-2101 Fax: (310) 206-5518 E-mail:

*Acknowledgments and Author Contributions: Data used in preparing this article were obtained from the

    Alzheimer's Disease Neuroimaging Initiative database ( Consequently, many ADNI investigators contributed to the design and implementation of ADNI or provided data but did not participate in the analysis or writing of this report. A complete listing of ADNI investigators is available at This work was primarily funded by the ADNI

    (Principal Investigator: Michael Weiner; NIH grant number U01 AG024904). ADNI is funded by the National Institute of Aging, the National Institute of Biomedical Imaging and Bioengineering (NIBIB), and the Foundation for the National Institutes of Health, through generous contributions from the following companies and organizations: Pfizer Inc., Wyeth Research, Bristol-Myers Squibb, Eli Lilly and Company, GlaxoSmithKline, Merck & Co. Inc., AstraZeneca AB, Novartis Pharmaceuticals Corporation, the Alzheimer‟s Association, Eisai Global Clinical

    Development, Elan Corporation plc, Forest Laboratories, and the Institute for the Study of Aging (ISOA), with participation from the U.S. Food and Drug Administration. The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer‟s Disease Cooperative Study at the University of California, San Diego. Algorithm development for this study was also funded by the NIA, NIBIB, the National Library of Medicine, and the National Center for Research Resources (AG016570, EB01651, LM05639, RR019771 to PT). Author contributions were as follows: XH, AL, NP, SL, MC, AT, and PT performed the image analyses; CJ and MW contributed substantially to the image acquisition, study design, quality control, calibration and pre-processing, databasing and image analysis. We thank Anders Dale for his contributions to the image pre-processing and the ADNI project. Part of this work was undertaken at UCLH/UCL, which received a proportion of funding from the Department of Health‟s NIHR Biomedical Research Centres funding scheme.

Abstract: (250 words; max: 250 words)

    In one of the largest brain MRI studies to date, we used tensor-based morphometry (TBM) to create 3D maps of structural atrophy in 676 subjects with Alzheimer‟s disease (AD),

    mild cognitive impairment (MCI), and healthy elderly controls, scanned as part of the Alzheimer‟s Disease Neuroimaging Initiative (ADNI). Using inverse-consistent 3D non-

    linear elastic image registration, we warped 676 individual brain MRI volumes to a population mean geometric template. Jacobian determinant maps were created, revealing the 3D profile of local volumetric expansion and compression. We compared the anatomical distribution of atrophy in 165 AD patients (age: 75.6 ? 7.6 years), 330 MCI subjects (74.8 ? 7.5), and 181 controls (75.9 ? 5.1). Brain atrophy in selected regions-of-interest was correlated with clinical measurements - the sum-of-boxes clinical dementia rating (CDR-SB), mini-mental state examination (MMSE), and the logical memory test scores - at voxel level followed by correction for multiple comparisons. Baseline temporal lobe atrophy correlated with current cognitive performance, future cognitive decline, and conversion from MCI to AD over the following year; it predicted future decline even in healthy subjects. Over half of the AD and MCI subjects carried the ApoE4 (apolipoprotein E4) gene, which increases risk for AD; they showed greater hippocampal and temporal lobe deficits than non-carriers. ApoE2 gene carriers - 1/6 of the normal group - showed reduced ventricular expansion, suggesting a protective effect. As an automated image analysis technique, TBM reveals 3D correlations between neuroimaging markers, genes, and future clinical changes, and is highly efficient for large-scale MRI studies.


    Alzheimer‟s disease (AD) is the most common form of dementia, affecting more than 5 million individuals in the U.S. alone, and over 24 million people worldwide. Although the exact time course is unknown, AD-related pathogenesis is believed to begin decades before clinical symptoms, such as memory impairment, can be detected {Price, 1999 #93; Goldman, 2001 #94; DeKosky, 2003 #88}. Several therapeutic trials now aim to resist disease progression in those with amnestic mild cognitive impairment (MCI) usually a

    transitional state between normal aging and AD - in which 10-25% of subjects develop AD within one year {Petersen, 1999 #9; Petersen, 2000 #7; Petersen, 2001 #8}. As AD develops, patients suffer from progressive decline in executive function, language, affect, and other cognitive and behavioral domains. To identify factors that accelerate or resist disease progression, such as treatment, genetic factors, and their interactions, it is imperative to develop biomarkers, or quantitative imaging measures, that can (1) detect abnormal aging before neuronal loss is widespread; (2) gauge the level of structural brain degeneration in way that correlates with standard cognitive measures, and (3) predict future clinical decline, or imminent conversion from MCI to AD {Mueller, 2005 #29}.

    Magnetic resonance imaging (MRI) is widely used in AD studies as it can non-invasively quantify gray and white matter integrity with high reproducibility {Leow, 2006 #88}.

    MRI-based measures of cortical and hippocampal atrophy have been used in recent clinical trials {Grundman, 2002 #105; Jack, 2003 #96}, and they have been shown to correlate with pathologically confirmed neuronal loss and with the molecular hallmarks of AD {Jack, 2002 #89; Silbert, 2003 #91}. There is interest in which MRI-based measures can optimally predict future clinical decline, often defined as conversion to AD over a specific follow-up interval {Jack, 1998 #95; Apostolova, 2006 #47; Scahill, 2003 #55; Fleisher, 2008 #97}, and which measures link best with standard cognitive assessments {Fox, 1999 #90; Jack, 2004 #92; Thompson, 2007 #46; Ridha, 2008 #114}.

    Statistical mapping methods have also been developed to quantify brain atrophy in 3D throughout the brain, offering a detailed perspective on the anatomical distribution of disease-related changes. Tensor-based morphometry (TBM) is one such image analysis technique that identifies regional structural differences from the gradients of the non-linear deformation fields that align, or „warp‟, images to a common anatomical template (reviewed in {Ashburner, 2003 #22}). At each voxel, a color-coded Jacobian determinant value indicates local volume excess or deficit relative to the corresponding anatomical structures in the template {Freeborough, 1998 #25; Fox, 2001 #27; Ashburner, 2003 #22; Chung, 2001 #27; Riddle, 2004 #24}. TBM provides a wide range of regional assessments from the voxel level to whole-brain analysis, and substructure volumes can be estimated simply by integrating Jacobian determinant values over a candidate region of interest. Since TBM requires little manual interaction, it has been recognized as a favorable technique for very large-scale brain MRI studies and as a candidate for use in clinical trials.

    In this study, we related MRI-derived TBM measurements to genetic, clinical and cognitive assessments made at the time of the scan and over the following year. We examined a large sample (N=676) to investigate genetic influences on the level of atrophy, and clinical correlations. Specifically, we were interested in correlating baseline temporal lobe atrophy, at the voxel level, with the sum-of-boxes clinical dementia rating (CDR-SB), change in CDR-SB over the following year, mini-mental state examination (MMSE), change in MMSE in the following year, the logical memory test (immediate and delayed), and conversion from MCI to AD. The goal was to determine specific regions in which atrophy predicts future decline. We also assessed ApoE gene effects on the level of atrophy, including not only the detrimental effect of the ApoE4 gene {Burggren, 2008 #127; Chou, 2008 #126; Chou, 2008 #125}, but also the hypothesized protective effects of the ApoE2 gene, which has been difficult to detect in small samples as its allelic frequency is relatively rare.

Materials and methods:


    Imaging data was analyzed from subjects scanned as part of the Alzheimer‟s Disease Neuroimaging Initiative (ADNI), a large five-year study launched in 2004 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 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 assessments acquired in a multi-site manner mirroring enrollment methods used in clinical trials, can replicate results from smaller single site studies measuring 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, M.D., VA Medical Center and University of California, San Francisco.

    676 baseline MRI scans, including 165 AD patients (age: 75.6 ? 7.6 years), 330 amnsetic MCI subjects (74.8 ? 7.5 years), and 181 healthy elderly controls (75.9 ? 5.1 years), were included in this study. All subjects underwent thorough clinical and cognitive assessment at the time of scan acquisition; scores are summarized in Table 1 (in the Results section).

    As part of each subject‟s cognitive evaluation, the clinical dementia rating (CDR) was used to measure dementia severity by evaluating patients‟ performance in six domains:

    memory, orientation, judgment and problem solving, community affairs, home and hobbies, and personal care {Hughes, 1982 #3; Berg, 1988 #122; Morris, 1993 #4}. We used the „sum-of-boxes‟ CDR scores, which have a dynamic range of 0-18; higher scores

    signify poorer cognitive function. The mini-mental state examination (MMSE) was administered to provide a global measure of mental status, evaluating five cognitive domains: orientation, registration, attention and calculation, recall, and language {Cockrell, 1988 #2; Folstein, 1975 #1}. The maximum MMSE score is 30; scores of 24 or lower are generally consistent with dementia. The logical memory (LM) test is a modified version of the episodic memory assessment from the Wechsler Memory Scale-Revised (WMS-R) {Wechsler, 1987 #123}. Subjects were asked to recall a short story that consists of 25 pieces of information, both immediately after it was read to the subject, and after a 30 minute delay. A maximum score is 25 with every recalled item of information accounting for 1 point. Approximately one year after baseline, subjects returned for a follow-up brain MRI scan and clinical assessment. Any changes in diagnosis were also noted. All AD patients met NINCDS/ADRDA criteria for probable AD {McKhann, 1984 #21}. ApoE genotyping was determined using DNA obtained from subjects‟ blood samples and was performed at the University of Pennsylvania. Please refer to the ADNI protocol for detailed inclusion and exclusion criteria {Mueller, 2005 #30; Mueller, 2005 #29}.

    This dataset was downloaded by February 1, 2008, and reflects the status of the database at that point; as data collection is ongoing, we focused on analyzing all available baseline scans, together with baseline and 1-year follow-up clinical and cognitive scores, as well as information on conversion from MCI to AD over the 1-year follow-up period. The study was conducted according to the Good Clinical Practice guidelines, the Declaration of Helsinki and U.S. 21 CFR Part 50-Protection of Human Subjects, and Part 56-Institutional Review Boards. Written informed consent was obtained from all participants before protocol-specific procedures, including cognitive testing, were performed.

MRI Acquisition and Image Correction

    As detailed elsewhere, all subjects were scanned with a standardized MRI protocol developed for ADNI {Leow, 2006 #88; Jack, 2008 #86}, summarized briefly here. High-resolution structural brain MRI scans were acquired at 58 ADNI sites using 1.5 Tesla MRI scanners (ADNI also collects a smaller subset of data at 3 Tesla but it was not analyzed here to avoid the additional complications of combining data across scanner field strengths). For each subject, two T1-weighted MRI scans were collected using a sagittal 3D MP-RAGE sequence. As described in a study by Jack et al. {Jack, 2008 #86} typical 1.5T acquisition parameters were repetition time (TR) of 2400 ms, minimum full TE, inversion time (TI) of 1000 ms, flip angle of 8?, 24 cm field of view, 192x192x166 acquisition matrix in the x-, y-, and z- dimensions, yielding a voxel size of 1.25x1.25x1.2 3mm. In plane, zero-filled reconstruction yielded a 256x256 matrix for a reconstructed 3voxel size of 0.9375x0.9375x1.2 mm. The images were calibrated with phantom-based

    geometric corrections to ensure consistency among scans acquired at different sites (Gunter et al., 2006).

    Image corrections were applied using a processing pipeline at the Mayo Clinic, consisting of: (1) a procedure termed GradWarp for correction of geometric distortion due to

    gradient non-linearity {Jovicich, 2006 #32}, (2) a “B1-correction”, to adjust for image

    intensity inhomogeneity due to B1 non-uniformity using calibration scans {Jack, 2008 #86}, (3) “N3” bias field correction, for reducing residual intensity inhomogeneity {Sled, 1998 #44}, and (4) geometrical scaling, according to a phantom scan acquired for each subject {Jack, 2008 #86}, to adjust for scanner- and session-specific calibration errors. In addition to the original uncorrected image files, images with all of these corrections already applied (GradWarp, B1, phantom scaling, and N3) are available to the general scientific community.

Image Pre-processing

    To adjust for global differences in position and scale across subjects, individual scans were linearly registered to the International Consortium for Brain Mapping template (ICBM-53) {Mazziotta, 2001 #47} using 9-parameter (9P) registration {Collins, 1994 #102}. Globally aligned images were resampled in an isotropic space of 220 voxels along 3each axis (x, y, and z) with a final voxel size of 1mm.

Unbiased Group Average Template - Minimal Deformation Target (MDT)

    A minimal deformation target (MDT) was created for the normal group to serve as an unbiased average template image {Good, 2001 #67; Joshi, 2004 #112; Kochunov, 2002 #65; Christensen, 2006 #113; Kovacevic, 2005 #66; Leporé, 2008 #127; Lorenzen, 2006 #103; Studholme, 2004 #114}. Using 40 randomly selected normal subjects, we created a customized template that facilitates automated image registration, reduces bias, and, in some studies, may improve statistical power {Leporé, 2007 #86}.

    The process of MDT construction was detailed previously {Hua, 2008 #124; Hua, 2007 #71} and is described briefly here. To construct an MDT, the first step was to create an initial affine average template, by taking a voxel-wise average of the 9P globally aligned scans after intensity normalization. In the second step, a non-linear average template was

    built after warping individual brain scans to the affine template. We utilized a non-linear inverse consistent elastic intensity-based registration algorithm {Leow, 2005 #23; Leow, 2005 #63} which optimizes a joint cost function based on mutual information (MI) and the elastic energy of the deformation. The deformation field was computed using a spectral method to implement the Cauchy-Navier elasticity operator {Marsden, 1983

    #122; Thompson, 2000 #93} using a Fast Fourier Transform (FFT) resolution of 32x32x32. This corresponds to an effective voxel size of 6.875 mm in the x, y, and z dimensions (220 mm / 32 = 6.875 mm). A non-linear average intensity template was then derived from the mean of the 40 deformed scans that had been non-linearly registered toward the affine average template. In a final step, the MDT was generated for the normal group by applying inverse geometric centering of the displacement fields to the non-linear average {Kochunov, 2002 #65; Kochunov, 2005 #69; Leporé , 2008 #128}.

Three-dimensional Jacobian Maps

    To quantify 3D patterns of volumetric tissue change, all individual brain images (N=676)

    were non-linearly aligned to the MDT for the normal group {Leow, 2005 #23}. For each subject, a separate Jacobian matrix field was derived from the gradients of the deformation field that aligned that individual brain to the MDT template. The determinant of the local Jacobian matrix was derived from the forward deformation field to characterize local volume differences. Color-coded Jacobian determinants were used to

    ;;detJr1illustrate regions of volume expansion, i.e. those with, or contraction, i.e.,

    ;;detJr1 {Ashburner, 2003 #22; Chung, 2001 #27; Freeborough, 1998 #25; Riddle, 2004 #24; Thompson, 2000 #28; Toga, 1999 #26} relative to the normal group template. As all images were registered to the same template, these Jacobian maps share a common anatomical coordinate defined by the normal template. Individual Jacobian maps were retained for further statistical analyses.

Pair-wise group comparisons

    Using TBM, we created 676 Jacobian maps that represent individual deviations from the normal brain template. To illustrate systematic changes between groups, we constructed voxel-wise statistical maps based on a Z statistic (this was used instead of a Student‟s t

    statistic, as the number of degrees of freedom was extremely high). We computed the overall significance of group differences using permutation tests, corrected for multiple comparisons {Bullmore, 1999 #61; Nichols, 2002 #76; Thompson, 2003 #52; Chiang, 2007 #77; Chiang, 2007 #81; }. In brief, a null distribution for the group differences in Jacobian at each voxel was constructed using 10,000 random permutations. For each test, the subjects‟ diagnosis was randomly permuted and voxel-wise Z tests were conducted to

    identify voxels more significant than p = 0.01. The volume of voxels in the brain more

    significant than p = 0.01 was computed for the real experiment and for the random assignments. A ratio, describing the fraction of the time the suprathreshold volume was greater in the randomized maps than the real effect (the original labeling), was calculated to give an overall P-value for the significance of the map. This procedure has been used in many prior reports {Chiang, 2007 #77; Hua, 2007 #71; Hua, 2008 #124}.

Regions of interest (ROIs)

    The MDT template was manually parcellated using the Brainsuite software program {Shattuck, 2002 #36} by a trained anatomist to generate binary masks for frontal, parietal, temporal, and occipital lobes. The hippocampus was delineated on the control average template by investigators at the University College London using the MIDAS (Medical Image Display and Analysis System) software {Freeborough, 1997 #57}. The delineation included hippocampus proper, dentate gyrus, subiculum, and alveus {Fox, 1996 #58; Scahill, 2003 #55}.

    Correlations of Structural Brain Differences (Jacobian Values) with Clinical Measurements

    At each voxel, correlations were assessed, using the general linear model, between the Jacobian values and several clinical measures the CDR-SB, change in CDR-SB over

    the following year, MMSE, change of MMSE over the following year, logical memory test (immediate and delayed), and conversion from MCI to AD during the following year. As Jacobian maps are composed of signals denoting both CSF expansion (J>1) and tissue loss (J<1), we performed separate evaluations of the positive, negative and two-sided associations. The results of voxel-wise correlations were corrected for multiple comparisons by permutation testing. In each random sample, clinical scores were randomly assigned to each subject and the number of voxels with significant correlations (p ? 0.01) was recorded. After 10,000 permutations, a ratio was calculated describing the fraction of the null simulations in which a statistical effect had occurred with similar or greater magnitude than the real effects. This ratio served as an estimate of the overall significance of the correlations, corrected for multiple comparisons, as performed in many prior studies {Nichols, 2002 #76}. The number of permutations N was chosen to be

    10,000, to control the standard error SEp of the omnibus probability p, which follows a

    binomial distribution B(N, p) with known standard error (Edgington, 1995). When N =

    10,000, the approximate margin of error (95% confidence interval) for p is around 5% of

    . p

Influence of Genetic Variants on Brain Structures

    To investigate how apolipoprotein E (ApoE) genotype modulates brain structure in each diagnostic group, we created age- and gender-matched groups for each diagnosis, categorized by their different combinations of ApoE alleles. Carriers of an ApoE2 gene (which confers lower risk for AD than that of the general population) or an ApoE4 gene (which increases risk for developing AD relative to that of the general population) were compared with homozygous ApoE3, which is the commonest genotype {Corder, 1993 #40; Saunders, 1993 #79; Roses, 1994 #78}. Group differences were quantified using voxel-wise Z-tests followed by permutations to correct for multiple comparisons, as described earlier.


     AGE CDR-SB MMSE Logical Memory Test

    (mean ? SD) Immediate Delay

    75.6 ? 7.6 4.41 ? 1.62 23.27 ? 2.03 3.99 ? 2.95 1.24 ? 1.83 AD

    74.8 ? 7.5 1.59 ? 0.86 27.03 ? 1.81 7.17 ? 3.25 3.86 ? 2.76 MCI

    75.9 ? 5.1 0.03 ? 0.11 29.13 ? 0.97 13.92 ? 3.39 13.01 ? 3.59 NORMAL

Table 1. Demographic and cognitive data for the subjects included in this study.

3D Profiles of Brain Atrophy

    We conducted pair-wise comparisons among AD (N = 165), MCI (N = 330) and normal

    (N = 181) groups. From individual Jacobian maps, we derived mean group difference maps and statistical maps based on the Z test. The 3D map comparing AD with normal

    controls revealed profound atrophy of the hippocampus and temporal lobes bilaterally in AD, accompanied by CSF expansion in the lateral ventricles and circular sulcus of the insula (Figure 1a). The MCI group displayed similar patterns of atrophy as AD but to a much lesser extent, with more anatomically restricted temporal lobe deficits (Figure 1b).

    Additional temporal lobe atrophy was apparent when contrasting the AD and MCI groups (Figure 1c). Permutation tests confirmed significant group differences in all three pair-wise comparisons using a whole-brain ROI, corrected for multiple comparisons. The two-tailed corrected P values were: P < 0.001 for AD vs. Normal, P < 0.001 for MCI vs. Normal, and P = 0.003 for AD vs. MCI.

Figure 1: 3D Maps of Brain Atrophy. The top rows of panels a, b, and c show the mean

    level of atrophy as a percentage reduction in volume. The bottom rows show the significance of these reductions, revealing highly significant atrophy in AD but a more anatomically restricted atrophic pattern in MCI. In MCI (b), atrophy is most prominent in the left hippocampus; as expected {Carmichael, 2007 #129; Carmichael, 2007 #130}, ventricular expansion is also substantial. When MCI is compared with AD (c), additional temporal lobe degeneration is evident. The overall level of ventricular expansion is about 10-15% for MCI and greater than 20% for the AD group.

Correlations with Clinical Measurements

    There is great interest in determining regions of the brain in which atrophy is most strongly correlated with established measures of cognitive or clinical decline, or with future outcome measures, such as imminent conversion to AD. These measures could potentially be used to inform prognosis or monitor drug treatment efficacy. Results from the temporal lobe ROI are shown below.

Sum-of-boxes CDR and future cognitive decline

    In both AD and MCI groups, the levels of baseline temporal lobe atrophy were correlated with the sum-of-boxes CDR scores. The negative correlations in Figure 2a identify

    regions in which tissue loss was linked with higher CDR-SB scores, i.e., more impaired cognitive function. The significance is slightly greater for the MCI than the AD group, perhaps partly due to the greater sample size for the MCI group. This test was not significant within the normal group, due to the low variance in CDR-SB scores. Additionally, we examined the correlation between the Jacobian values (which reflect structure volumes at baseline) and the changes in CDR-SB over the following year (Figure 2b). In all three groups, independently, there was a significant negative correlation, suggesting that baseline temporal atrophy predicts future cognitive decline.

Report this document

For any questions or suggestions please email