Computational anatomic methods as applied to aging and
dementia
1,2
PAUL M THOMPSON,
PhD
and
1,2LIANA G APOSTOLOVA,
MD1Laboratory of NeuroImaging and2Department of Neurology, UCLA School of Medicine, Los Angeles, CA, USA
ABSTRACT.The cellular hallmarks of Alzheimer’s disease (AD) accumulate in the living brain up to 30 years before the characteristic symptoms of dementia can be identified. Brain changes in AD are difficult to distinguish from those in normal aging, and this has led to the development of powerful computational methods to extract statistical information on the brain changes that are characteristic of AD, mild cognitive impairment (MCI), and different dementia subtypes. Time-lapse maps can be built to show how the disease spreads in the brain, and where treatment affects the disease trajectory. Here, we review three computational approaches to map brain deficits in AD: cortical thickness maps, tensor-based morphometry, and hippocampal/ventricular surface modeling. Anatomical structures, modeled as three-dimensional geometrical surfaces, are mathematically combined across subjects for group or interval
comparisons. Mathematical concepts from computational surface modeling, fluid mechanics, and multivariate statistics are exploited to distinguish disease from normal variations in brain structure. These methods yield insight into the dynamics of AD and MCI, showing where brain changes correlate with cognitive or behavioral changes such as language dysfunction or apathy. We describe cortical and hippocampal changes that distinguish dementia subtypes (such as Lewy-body dementia, HIV-associated dementia and AD), and we describe brain changes that predict recovery or decline in those at risk. Finally, we indicate which computational methods are powerful enough to track dementia in clinical trials, on the basis of their efficiency and sensitivity to early change, and the detail in the measures they provide.
Received 9 August 2007 Revised 29 December 2007 Accepted 24 January 2008 DOI: 10.1259/BJR/20005470 ’2008 The British Institute of Radiology
Classical maps of the spread of Alzheimer’s disease (AD) pathology in the brain [1] present a challenge for all those using imaging to understand dementia. These maps of neurofibrillary tangles and plaques, painstak-ingly compiled from histologically stained brain sections, give a picture of a the advance of disease years beforein vivoimaging could detect any sign of pathology. In Braak neurofibrillary stages III and IV, which correspond to the onset of AD, severe damage has already occurred in the entorhinal cortex and neighboring areas. Disturbingly, this undetectable advance of AD is proceeding in billions of individuals alive today, with no means to protect them or even diagnose the illness before irreversible neuronal loss has set in.
Major advances are occurring in scanning techniques and in image analysis, making it easier to track disease progression with greater power. In studies of hundreds of subjects scanned over time, databases of images can now be combined to visualize the disease trajectory, showing the spread of cortical atrophy, impaired metabolism, and even plaque and tangle accumulation. Ultra-high-field MRI reveals early changes in specific hippocampal subfields [2, 3], and new positron emission tomography (PET) ligands are emerging that make it possible to visualize plaque and tangle pathology [4–7].
Diffusion tensor imaging is also advancing to ever-higher magnetic fields, mapping white matter degenera-tion with better signal to noise and anatomical precision [8].
Even with the best imaging techniques available today, however, advanced image-analysis methods are still required to detect brain changes that predict prognosis in mild cognitive impairment (MCI) or AD, or to show that therapy is effective [9, 10]. In clinical trials that use imaging, often hundreds to thousands of subjects are scanned to detect systematic differences in rates of atrophy [11].
Several of these imaging-based trials have been successful [12, 13], and there is great interest in detecting treatment effects after short follow-up intervals (e.g., 3–6 months) [14]. Given the need to detect treatment effects as early as possible, there has been pressure to improve not just the image resolution and contrast but also the computational methods needed to resolve systematic effects [9]. Here, we review some recent advances in these computational methods, focusing on techniques that have yielded new information on brain changes in aging and dementia.
Computational anatomy
Computational anatomy is a general term covering a broad class of mathematical methods that model
Address correspondence to: Dr Paul Thompson, Professor of Neurology, Laboratory of Neuro Imaging, Department of Neurology, UCLA School of Medicine, 635 Charles Young Drive, Los Angeles, CA 90095-1769, USA. E-mail: [email protected]
anatomical structures in images as three-dimensional (3D) curves, surfaces and maps, and which combine them across subjects to create statistical maps of disease. Computer algorithms can combine information from many hundreds of subjects to examine the statistics of features such as cortical gray matter thickness [15, 16], functional MRI (fMRI) activation [17], metabolism [18] or molecular pathology [6]. To pool data from many subjects, sophisticated registration approaches are used to align image datasets to a common coordinate space. This alignment is often based on computationally guided matching of cortical features such as gyral/sulcal land-marks, identified either manually or using computer-vision approaches [19]. Statistics can then be computed to detect subtle brain changes that are associated with prognosis or treatment (or other factors of interest), making it possible to visualize group differences in 3D as a statistical map.
Spatially detailed maps of atrophy are useful in several contexts. First, they reveal brain regions that change earliest, or most significantly, at a specific stage of disease. For example, they can show profiles of brain changes that typically occur as subjects convert from normal cognition to MCI [20] and from MCI to AD [21, 22]. Second, different patterns of atrophy characterize different types of dementia, such as fronto-temporal versus Lewy-body dementia, and understanding them can assist with differential diagnosis. Third, longitudinal maps from AD populations and healthy controls can be compared to create time-lapse movies of disease pro-gression [11, 16, 20]. This provides insight into the sequence of cognitive impairment in AD. Time-lapse maps identify brain regions that are relatively resistant to pathology. The sensorimotor cortices, for example, mature in early infancy but degenerate only late in AD [23]. Statistical data can then be compiled to test which factors affect neurodegeneration, and where in the brain the disease is slowed. Time-lapse maps can also establish whether cortical thinning and amyloid plaque accumula-tion are found in the same brain regions at the same times, shedding light on the cascade of cellular events that lead to AD. Also, given the confounding effects of anatomical atrophy on PET and other metabolic or functional signals, multi-modality analysis tools are beginning to relate point-by-point changes in amyloid load to cortical thinning on MRI, and to disentangle the effects of one signal from the other [6].
Cortical thickness mapping
The most prominent features of AD-related neurode-generation on a standard T1-weighted MRI scan are
progressive sulcal and cerebrospinal fluid (CSF) space enlargement, and diffuse cortical atrophy. Gray matter atrophy spreads in a stereotypical sequence: from the hippocampus to the rest of the limbic system, and eventually to the rest of the neocortex [1]. Cellular shrinkage, neuropil loss and intracortical myelin reduc-tion [24, 25] all lead to age- and disease-related cortical thinning, which is visible, even at the gross anatomic level, on MRI. To quantify the amount of tissue loss precisely, overall volumes of gray and white matter can be computed from brain MRI scans. Tissue classification
methods can assign each image voxel to a specific tissue class (see Figure 1, 2ndrow, for an example). These tissue classifiers often use sophisticated Bayesian methods to fit statistical models to the MRI signal intensities in a scan, while adjusting for spatial intensity in homogeneities that result from non-uniformities in the scanner magnetic field. They also take into account the statistical likelihood of finding each tissue type at each location in a stereotaxic space [26].
If gray matter maps from many subjects are aligned to a standard 3D digital brain atlas, on which lobar subdivisions have been labeled, regional volumes for each tissue type can be derived by counting voxels. These region-of-interest measures, in conjunction with manual tracings of the hippocampus and other sub-cortical structures, have been the mainstay of morpho-metric analysis for two decades. Additional processing makes it possible to create spatial maps of more detailed anatomical measures, such as cortical gray matter thickness, for cross-subject and group comparisons. Gray matter thickness maps, in particular, can detect atrophy more powerfully than simple volumetric mea-sures, especially when disease effects do not perfectly coincide with a pre-specified region traced on the image. For instance, in AD, the only cortical subregions where changes correlate strongly with apathy are the cingulate gyrus and supplementary motor cortex (Figure 3b). These regions are readily seen in a map but may be overlooked if tissue counts from each lobe are examined [27]. Maps can also localize disease effects in 3D relative to sulcal/gyral landmarks. Specialized map-based sta-tistics, based on Gaussian field theory or false discovery rate, can powerfully detect disease effects, and they can be tuned to provide greater power depending on whether a signal is localized or weak and diffuse over much of the cortex [28].
One approach for cortical thickness mapping is presented in Figure 1, which shows a computational pipeline for processing MRI scans [16]. In each scan, a 3D map is first computed to quantify the distance of cortical gray matter voxels to the gray/white matter interface. Thickness values, in millimeters, are calculated using the Eikonal fire equation with fully 3D front propagation from the inner cortical surface to the outer cortical surface, and are plotted at each point on a cortical surface model extracted from the scan. To combine thickness data across subjects (Figure 1, 3rd row), a spherical coordinate system is imposed onto each subject’s cortical surface. This serves as a reference grid so that thickness data can be compared at a given surface-based coordi-nate across subjects. If sulcal/gyral landmarks are traced onto the cortical models, data from corresponding cortical regions can be averaged across subjects using a technique known as cortical pattern matching [16]. The sulcal pattern is digitized, and using a flow field in spherical coordinates, cortical landmarks can be exactly matched across subjects. This is useful for two reasons. First, an average cortical model can be created for a group of subjects (such as an AD or MCI population), with well-defined sulcal features in their mean spatial locations (Figure 1, 5throw). Second, maps of thickness
values in each individual can be transferred onto this average model while adjusting for the individual gyral patterning differences across subjects. This method
COLOR
FIGURE
allows pixel-by-pixel averaging of values within each delineated region across all subjects, allowing compar-isons of patients with controls. Averaging these aligned cortical data can also reinforce consistent features. Now that each subject’s thickness data are aligned to a common space, a statistical model can be fitted to the thickness measures from all subjects at each surface point (Figure 1, bottom row). In brain mapping studies, it is common to fit statistical models to data at each location in an image to assess effects of age, diagnosis or experimental parameters (an approach known as statis-tical parametric mapping [SPM]; [29]). The results are displayed in the form of a map of statistics, and the overall significance of the map can be assessed using formulae for the distribution of features in Gaussian random fields, or by using permutation methods, which randomly assign subjects to groups to find out how likely effects are to occur by accident. When statistics are computed for data on surfaces, regions of significance can be color-coded to show the distribution of atrophy [16].
Cortical thickness mapping is related to two simpler but widely used methods. Before algorithms became widely available for computing cortical thickness accu-rately from MRI scans [16, 30–32], it was more common to compute a local measure of gray matter volume called ‘gray matter density’ (GMD). This is defined, at each point in an image, as the proportion of tissue segmenting as gray matter in a small spherical region (typically of 10–12 mm radius) centered at that point [33, 34]. GMD and thickness are highly correlated [35], except at the temporal lobe tips where cortical curvature is high. GMD is easy to measure as it does not require accurate modeling of the inner and outer cortical surfaces in each scan. Owing to the spatial smoothing implicit in its definition, GMD is quite robust to image noise. One popular method, voxel-based morphometry (VBM; [36, 37]), compares maps of smoothed GMD voxel-by-voxel after spatially normalizing all datasets to the same coordinate space. Initial critiques of VBM focused on concerns that it might detect spurious group differences caused by misregistration of data into the common space, or because of interactions between diagnosis and registration errors [38, 39]. More recently, VBM has been improved, as have the methods to align brains into a common space, adjusting for complex shape differences [40, 41]. Although registration seems like a technical detail, it influences power to detect atrophy in a multi-subject study as any anatomical misregistration sacrifices signal-to-noise. Cortical-pattern matching improves registration by matching anatomical landmarks, remov-ing confoundremov-ing variance due to the mismatch of cortical
anatomy across subjects. Because sulcal landmarking is time-consuming, some groups have attempted auto-mated matching of cortical features by matching mean curvature maps using information theory [42] or by attempting to find cortical sulci automatically [43]. Additional efforts to combine surface and fully 3D volumetric registration are underway [44], avoiding the need to normalize cortical gray matter and white matter data using different specialized approaches that are not yet compatible with each other [45].
Cortical maps in AD and MCI
Cortical mapping methods have distinguished pat-terns of atrophy that are typical of late- versus early-onset AD [47], and different dementia subtypes [48]. They have also identified cortical changes associated with cognitive deterioration [49, 50]. Sowell et al. [51] applied cortical pattern matching to 176 scans of subjects aged 7 to 87, and compiled trajectories of gray matter thinning for each cortical point over the lifespan. Encouragingly, there was close regional correspondence between the cortical thickness maps created fromin vivo
MRI and the 1929post mortemdata of von Economo. The
;
MRI measures show massive attrition of frontal gray matter in late adolescence, but it was only late in life that a downswing in the amount of temporal gray matter occurred. The cortex does not age in a homogeneous way; each cortical region has a somewhat distinctive trajectory [51, 52]. A related study by Gogtay et al. [23] created a time-lapse map of cortical development from ages 4 to 22, and showed the earliest maturing cortex is least vulnerable to aging and AD (see http://www. loni.ucla.edu/,thompson/DEVEL/dynamic.html for maps). This phenomenon, illustrated in Figure 2, is sometimes referred to as retrogenesis [53]. As is visually evident in the time-lapse maps, the maturational sequence proceeds in a pattern opposite to the classical neurodegenerative sequence in AD.Apostolova et al. [46] compared gray matter profiles in 26 amnestic MCI and 31 mild AD subjects, showing strikingly greater atrophy in mild AD (Figure 3a) that conforms to the pattern of spread of AD pathology observedpost mortemthrough the brain. A related VBM study found significantly greater parietal, anterior and posterior cingulate atrophy in mild to severe AD relative to MCI [54].
With several promising disease-modifying candidate compounds under development, there is significant value in tracking these changes with anatomical preci-sion in an effort to show structural disease-modifying effects that influence the transition from MCI to AD.
Figure 1. Analyzing cortical degeneration. This flowchart shows an image analysis pipeline (adapted from Thompson et al., 2004) that can map degenerative changes in the cortex, such as gray matter atrophy. Cortical thickness profiles, for example, can be averaged across subjects or compared across time. Images are aligned to a standard brain template, such as an averaged MR image from a population (row 1). Maps of gray and white matter are mathematically coded to produce cortical thickness maps (row 2). Sulcal landmarks traced on extracted surface models (row 3) can serve as anchors that guide a flow field matching gyral regions across subjects (row 4). Then maps of cortical thickness, or other cortical signals (such as co-registered PET images), can be fluidly aligned to an average cortical model (row 5). Statistical models are fitted to data from all the subjects at each cortical point. These can assess whether imaging signals are associated with age (bottom row), diagnosis, cognition or other parameters of interest (e.g., medication or genotype). Effects of aging on cortical measures, or changes relative to baseline, may be animated as a time-lapse film to reveal the disease trajectory (see text for details).
Cortical thinning correlates highly with global mea-sures of cognitive decline, at least in MCI and AD. This makes it an excellent biomarker of neurodegeneration. One widely used global cognitive measure, the mini-mental state exam (MMSE) score, depends on the integrity of widely distributed cortical areas in both brain hemispheres with left-sided predominance [55, 56]. This dependency is clear, based on very widespread correlations between cortical thickness and MMSE scores.
Apostolova et al. [27] examined the structural corre-lates of apathy in AD by applying cortical mapping to 35 AD patients with and without apathy (Figure 3b). Apathy severity was associated with cortical gray matter atrophy in bilateral anterior cingulate (Brodmann area [BA] 24; r50.39–0.42, p50.01) and left medial frontal cortex (BA 8 and 9;r50.4,p,0.02). A subsequent study found associations between Boston Naming and the animal fluency tests (i.e., measures of language function) and cortical atrophy in 19 probable AD and 5 multiple domain amnestic MCI patients who later converted to AD [50]. The degree of language impairment correlated with cortical atrophy in the left temporal and parietal lobes, especially in the perisylvian language cortices. This regional correlation is of interest because there was
Figure 2. Degenerative sequence in AD is the reverse of the normal developmental sequence. In a process termed retro-genesis (e.g., by Reisberg et al., 1999), cortical regions that mature earliest in infancy tend to degenerate last in AD. The developmental sequence echoes the phylogenetic sequence in which structures evolved. The most heavily myelinated structures, with least neuronal plasticity, resist AD-related neurodegeneration. Arrows denote the childhood cortical maturation sequence (left panel; Gogtay et al., 2004) and the gray matter atrophy sequence in AD (right panel; Thompson et al., 2003). Images are from time-lapse films compiled from cortical models in subjects scanned longitudinally with MRI, which may be viewed at: http://www.loni.ucla.edu/, thomp-son/DEVEL/dynamic.html and http://www.loni.ucla.edu/ ,thompson/AD_4D/dynamic.html
Figure 3. (a) Cortical thinning of up to 20% occurs as MCI progresses to AD, in widely distributed cortical areas (areas with greatest tissue loss are colored red). (b) Gray matter atrophy in the anterior cingulate and supplementary motor cortices is correlated with the presence or absence of apathy in AD. Regions where structural variation correlates with clinical or behavioral differ-ences can be identified using corre-lation maps such as these (strong associations are shown here in red colors). (Data adapted from [27, 46].)
COLOR
FIGURE
COLOR
FIGURE
only a weak correlation between the language tests and overall cognitive decline as indexed by MMSE scores. Lu et al. [57] found that childhood cortical maturation (indexed by gray matter thinning) in language and motor cortices was associated with performance in language and motor functions, respectively but not vice versa. Regional cortical thinning may be associated specifically with differences in cognitive functions that recruit those cortical areas, in both normal brain development and in dementia.
In related study, Thompson et al. [58] fitted genetic models to cortical gray matter maps in 40 twins (10 mono- and 10 dizygotic pairs), and discovered that GMD was almost completely genetically determined and correlated with full-scale IQ. These findings were confirmed and extended by Posthuma et al. [59] and by Haier et al. [60], suggesting that the cognitive reserve that protects individuals from the advance of AD may be determined by genes that influence cortical develop-ment.
Frisoni et al. [47] compared cortical atrophy in early-and late-onset Alzheimer’s disease (EOAD versus LOAD) and found that EOAD affected neocortical and LOAD medial temporal areas, compared with age-matched controls. Consistent with neuropsychological findings of greater memory impairment in LOAD and impaired neocortical functions in EOAD [61], these different atrophy patterns suggest different predisposing or etiologic factors. At comparable levels of cognitive decline, total gray matter loss was 19.5% in EOAD but only about half as much (11.9%) in LOAD, relative to appropriately matched controls [47].
Time-lapse maps and the trajectory of AD
The cortical pattern matching approach may be extended to time-varying or functional imaging data. If longitudinal scans are available, a time-lapse movie of disease progression can be created by fitting a trajectory to cortical thickness, or any other imaging parameter, at each cortical point (Figure 1, bottom right). This is done by substituting the subject’s age, or the time elapsed since their baseline scan, back into the statistical model estimated from the cortical measures in the entire sample. A movie ‘‘frame’’ can then be written out for each time-point and the series of frames can be animated. An example of this type of dynamic map may be viewed at: http://www.loni.ucla.edu/,thompson/AD_4D/ dynamic.html. Here, a time-dependent model was fitted to cortical gray matter density in 14 AD patients and 12 controls scanned longitudinally for 4 years. Maps of the degree of deficits, either as a percentage or as a significance map, show that the cortex is thinner in medial and lateral temporal lobes in early AD and that deficits advance anteriorly to engulf the cingulate and frontal cortex. As noted earlier, primary sensorimotor cortices are spared until late in the disease. Figure 4 shows two time-points from this animation, showing that cortical atrophy on MRI proceeds in approximately the same anatomical sequence as plaque and tangle burden in histopathological studies of AD. Plaque and tangle deposition starts in medial temporal regions ([1]; top row of Figure 4) and affects the posterior limbic
system first because of its close connections to the posterior cingulate gyrus [62]. Hypometabolism of the posterior cingulate is observed early in AD, even when no atrophy is detected in this region [63]. On the other hand, in frontotemporal lobar degeneration, neuronal loss is first observed in the frontal regions closely connected to the anterior cingulate cortex [64].
Figure 4. Progression of AD based on pathology, MRI and [18F] FDDNP PET. Neurofibrillary tangles, one of the mole-cular hallmarks of AD, spread in the brain in a characteristic advancing trajectory (top row; adapted from Braak and Braak, 1991). Darker red colors denote areas with greater tangle deposition, based on histologically stained post mortemmaterial. On MRI, the areas with gray matter deficits in mild AD include primarily the temporal lobe, but in moderate AD, these deficits spread to involve the frontal cortex (middle row; adapted from a longitudinal study by Thompson et al., 2003). Finally, estimates of cerebral amyloid in vivo obtained using the PET ligand FDDNP are low in controls but higher in those with impaired cognition (bottom row; adapted from Braskie et al., 2007). The anatomical agreement is striking between these in vivo maps and the well-established post mortem maps for the staging of AD. In all maps, the sensorimotor cortex shows least disease-related degeneration. (Adapted, with permis-sion from the authors and publishers.)
COLOR
FIGURE
To better understand how amyloid load spreads in the living brain, Braskie et al. [6] applied cortical pattern matching to 23 subjects (10 controls, 6 amnestic MCI and 7 AD) scanned with both MRI and [18F]FDDNP, a recently developed PET ligand that is sensitive to plaque and tangle pathology [5]. They aligned parametric PET images of amyloid load to MRI scans from the same subjects, textured the PET signals onto the cortex, and combined them across subjects using cortical pattern matching ([6]; Figure 1, right column). Figure 4 shows two frames from an animation sequence that shows the degree of amyloid burden for different levels of cognitive impairment. The advancing pathology also follows the classical Braak trajectory for neurofibrillary tangle accumulation.
Related work by Mintun et al. [7] and by Rowe et al. [65] with Pittsburgh Compound B ([11C]PIB) shows frontal lobe labeling early in the degenerative sequence. The PIB progression pattern is consistent with the Braak trajectory for amyloid deposition, which, unlike tangle deposition, shows early increases in the basal neocortex, particularly in frontal and temporal lobes and primarily in poorly myelinated regions such as the perirhinal cortex. These PET changes may occur at a much earlier stage of the disease than cortical thinning; amyloid PET appears to be sensitive to pathological changes earlier than measures of gray matter, and is also correlated with subclinical cognitive decline even in normal subjects [6, 66]. Clearly the detectability of changes in each imaging modality depends on the population studied: the sample size as well as details of signal reconstruction, partial volume correction, and registration. As such, it is difficult to make absolute statements as to how the trajectory of various PET ligands (FDDNP versus PIB or others) relate chronologically to each other and to cortical thinning, unless all measures are compared head-to-head in the same subjects, which has not been done to date.
Even so, amyloid-sensitive PET signal correlates with cognitive performance even within the normal range, with correlations in cortical areas that deteriorate earliest in AD suggesting its potential utility for early diagnosis.
Tensor-based morphometry
While the methods described so far provide detailed maps of changes in cortical measures, tensor-based morphometry (TBM) can track volumetric changes throughout the brain. TBM can track longitudinal changes in 3D, such as rates of atrophy over time (in percent per year). Alternatively, in a cross-sectional design, it can compare baseline scans of two different groups, such as AD versus control groups.
Figure 5 shows the premise of the TBM approach. If a pair of scans is collected from the same subject over time, they can be aligned with each other, using a fluid transformation that applies compressions and expan-sions at a local level throughout the anatomy. One class of registration methods models the baseline image as a deformable elastic medium [67] or as a viscous fluid [40], and applies distributed internal forces to reconfigure the earlier scan to match the later one. So long as the matching is accurate, the spatial gradient of the transformation (‘deformed grid’ in Figure 5) measures how much tissue is lost over the time interval between the scans, and this can be plotted and color-coded. Applied to a sequence of scans acquired over time from the same patient, these voxel compression maps [68], also known as Jacobian maps [69], can reveal the extent and spread of atrophy. They are also amenable to voxel-by-voxel averaging [70], to making a map of the average atrophic rate in a group, or to the identification of regions where atrophic rates depend on some external parameter (which could be treatment, genotype, or future decline).
Figure 5. Tensor-based morphometry (TBM). In TBM, a fluid transformation (deformed grid) is applied to a baseline scan (Source) to reconfigure it (Deformed Source) into the shape of a follow-up scan (Target). The local expansions (red colors in (b)) can be plotted in brain regions such as the corpus callosum (shown in the green box in (a)), indicating regions with fastest growth (b) or atrophy (c). Voxel expansions (red colors) or contractions (blue colors) can be plotted onto a sequence of scans collected from the same subject during a degenerative brain disease, emphasizing regions with progressive atrophy. These maps may be averaged across subjects or compared across populations to assess factors that influence degenerative rates in each region of the brain.
COLOR
FIGURE
If only a single scan is available per subject, all scans may be fluidly aligned to a common anatomical template; in such a cross-sectional design, the degree of compression then reflects volume differences between each subject and the template, and maps of tissue loss can be made. Figure 6 shows TBM-based maps of atrophy in HIV/AIDS patients, along with maps of cortical thinning. Strikingly, the underlying white matter shows 10–20% volume reductions immediately below cortical regions with greatest thinning [40, 71]. This suggests that TBM and cortical thickness measures agree, although they are computed using very different image processing pipelines. Further analysis revealed that the level of deficits, in both cortex and white matter, correlated strongly with declining immunity (CD4+ T-cell counts), but not with viral load or anti-retroviral treatment, suggesting that such treatments have diffi-culty crossing the blood–brain barrier, even when they are effective in bolstering the immune system and prolonging life.
For large clinical studies, it is important to be aware of several factors that affect the power of TBM. First, anatomical features are matched by deforming one scan onto another, using a flow field that is adjusted to optimize a mathematical measure of image similarity between the two scans. Some ingenuity has gone into developing measures that can align images without interactive identification of landmarks (as in the cortical pattern matching method). Atrophy may be detected
most powerfully by image registration measures based on information theory, such as the Jensen-Re´nyi diver-gence, Kullback-Leibler divergence or Chernoff distance, and the relative power of these measures is being investigated [72, 73]. Second, if the full deformation tensor is analyzed at each voxel, rather than just the compression factor, the power to detect atrophy is vastly increased [67]. Power is also slightly increased if all brains are aligned to a common brain template that is representative of the mean anatomy of all subjects in the population, defined using Riemannian manifolds [74]. Third, rather than modeling the flow using continuum mechanics, which may not accurately reflect the atrophic behavior of the brain, deformation models can be bolstered by using Green’s functions to encode known patterns of spatial covariance in changes occurring in multiple brain regions [75] or directional anisotropy in the observed changes [76]. Given the pace of these mathematical developments, TBM is attractive for clinical trials because it allows large samples to be studied with no manual interaction [41, 77].
Hippocampal and ventricular radial mapping
Even though TBM can map patterns of atrophic changes throughout the brain, the fact remains that the medial temporal lobe is the site of the earliest structural change in AD. The hippocampus, in particular, is the
Figure 6. Visualizing brain tissue loss in HIV/AIDS. (Top row) In an MRI study of cortical thickness in 27 HIV/AIDS patients and 14 healthy controls, the primary sensory, motor and premotor cortices were an aver-age of 15% thinner in the patients; prefrontal and parietal tissue loss correlated with cognitive and motor deficits. Thinner frontopolar and language cortex also correlated with immune system deterioration mea-sured via blood levels of CD4+ T-lymphocytes. (Bottom row) When the same subjects were studied using tensor-based morphometry (Chiang et al., 2006 ), the pattern
FO
of white matter loss was in remark-able agreement with the cortical maps. In diseased brains, the white matter volume was reduced in the premotor areas, where the cortex was significantly thinner, suggesting that cortical degeneration may be accompanied by degeneration in the underlying white matter path-ways. Taken together, these and other studies support the notion that brain degeneration is present even in apparently healthy HIV-positive people on powerful drug regimens (i.e., highly active anti-retroviral therapy [HAART]). (Data illustrated in the top row are from Thompson et al., 2005 ; data illu-FP
strated in the bottom row are from Chiang et al., 2006.)FQ
COLOR
target of several specialized computational modeling methods. Some of these model the hippocampus as a 3D tube-like surface, from which measures of volume can be derived. In normal aging, hippocampal volume loss is on average 1.6–1.7% per year [78, 79], whereas that of the entorhinal cortex is ,1.4% per year [80]. Much higher rates of hippocampal volume loss are observed in MCI and AD, with faster atrophic rates in MCI subjects who decline to AD relative to those who remain stable (annual hippocampal atrophy rate for MCI patients who remain stable is 2.8%, for MCI converters 3.7% and for those with AD 3.5–4%; [81]).
In an effort to localize these changes to specific hippocampal sectors, Thompson et al. [82] proposed a radial atrophy mapping (RAM) approach which creates surface models to represent the hippocampus, imposes a regular grid structure on anatomical models from different individuals, and uses this structure to compute average shape models for different groups. As a local
index of atrophy, the distance of each surface point to a centerline threading down the center of the structure is plotted on the surface (Figure 7). Surface-based statistics on this measure can then be used to identify regions where atrophy is associated with diagnosis, cognition, genotype, or medication. Apostolova et al. [21] used this radial atrophy mapping approach to study 20 MCI subjects who were followed clinically and neuropsycho-logically for 3 years. Over the three-year period, 6 patients developed AD, 7 remained stable, and 7 improved. As shown in Figure 7 (bottom row), smaller hippocampi, and specifically CA1 and subicular involve-ment, were associated with increased risk of conversion from MCI to AD. MCI patients who improved and no longer met MCI criteria at follow-up tended to have larger hippocampal volumes, and their subiculum and CA1 regions were relatively preserved.
Using the same method, Becker et al. [83] showed greater hippocampal atrophy in amnestic versus
non-Figure 7. Mapping hippocampal atrophy. The radial atrophy mapping method (top row) relies on (A) manually or automatically tracing the hippocampus, (B) computing a 3D parametric mesh model of the structure, (C) estimating the distance between the central core of the structure to each surface point (i.e., the radial distance), and recording radial distance estimates at each surface point to create (D) individual and (E) group average color-coded maps of the radial distance. These maps are then statistically compared between groups or conditions. One longitudinal 3D MRI study [21] compared baseline hippocampal atrophy in MCI subjects who converted to AD with that in MCI subjects who remained stable or improved cognitively during 3 years of clinical follow-up. Greater atrophy at baseline was seen in the CA1 and subicular areas in subjects who subsequently converted to clinically probable AD (bottom right panels). The bottom panels show a schematic representation of the hippocampal subfields (left), the proportional difference in atrophy levels between converters and non-converters (in %; middle panels), and the significance of these differences taking into account normal structural variation (bottom right panel).
FR
COLOR
FIGURE
amnestic MCI and AD. When a group of 28 AD patients were compared with 40 cognitively intact persons, Frisoni et al. [84] found significant atrophic changes in the AD patients amounting to tissue loss of 20% or more in regions of the hippocampal formation corresponding to the CA1 field and part of the subiculum, but regions corresponding to the CA2-3 fields were remarkably spared.
Correlations with symptoms have also been examined using hippocampal maps. Ballmaier et al. [85] mapped atrophy of the hippocampal head in depressed versus non-depressed elderly subjects. Statistical mapping results, confirmed by permutation testing, showed that regional surface contractions in the anterior aspects of the subiculum were greater in elderly subjects with late-onset depression compared to those with early-late-onset depression, the same was true for regional surface contractions in lateral-posterior aspects of the CA1 subfield in the left hemisphere. Hippocampal surface contractions correlated with memory measures among late-onset depressed patients. In a study of patients with Lewy-body dementia (LBD), thought to be the second most common cause of degenerative dementia after Alzheimer’s disease, Sabattoli et al. [86] found hippo-campal atrophy in a milder and different pattern than that typical of AD. Typically, AD is characterized by early and progressive memory impairment. In LBD patients, memory disturbance appears later but atten-tional and executive impairment is prominent, consistent with the greater extent of frontal lobe atrophy mapped in LBD patients by cortical pattern matching methods. In LBD patients, hippocampal tissue was mainly lost in the anterior sector of CA1, with greater involvement of the hippocampal head, whereas AD patients were more atrophic in the posterior sector of CA1. Differences in the pattern of temporal atrophy between LBD and AD patients may help to explain differences in their cognitive presentation. Similar studies of patients with vascular dementia showed a more restricted pattern of hippocampal atrophy than seen in AD patients at a comparable level of cognitive impairment [87]. Hippocampal maps may therefore complement cortical maps in revealing the selective atrophic patterns that characterize different types of dementia.
Several factors affect the power and utility of hippo-campal maps. First, any manual segmentation slows down the throughput of the approach. The hippocampus may be segmented automatically via fluid registration [88–90], Markov random field methods [91], or computer vision approaches such as adaptive boosting [92]. Hippocampal mapping may be combined with auto-mated segmentations [92], but some investigators prefer manual tracing, which can be more accurate, as hippocampal geometry is so complex. Second, as higher-field imaging resolves internal subfields within the hippocampus [2], these subfields could be aligned in population studies to reinforce consistent features, along the lines of cortical-pattern matching, which tries to align corresponding features in the cortex. Approaches do exist for aligning hippocampal landmarks internal to the flattened cortical sheet [17, 93], but they are not used widely because subfields are not readily identifiable on conventionalT1-weighted images at 1.5 or 3 Tesla.
Finally, ventricular surface modeling provides a rapid, if non-specific, method for tracking the progression of dementia. Carmichael et al. [94, 95] used a deformable registration approach to align an image template, on which the ventricles had been labeled by hand, to 339 brain images from the Cardiovascular Health Study. They mapped group differences between AD, MCI and controls using the radial atrophy mapping approach of Thompson et al. [82]. The ventricles were split up into frontal, inferior and occipital horns, and radial distances were computed to index disease-related ventricular expansion. A related approach known as multi-atlas fluid image alignment (MAFIA) was developed by Chou et al. [96, 97], who fluidly registered multiple (up to nine) labeled ventricular surfaces to each individual scan in a database, substantially reducing segmentation errors by averaging multiple independent segmentations within each scan. These automated methods revealed that ventricular changes associated with the transition from MCI to AD occurred largely in the frontal horn; in healthy controls, localized ventricular expansions were associated with carrying the ApoE4 risk gene or AD. The very high automation of this approach recommends it for use in clinical trials and epidemiological studies.
Practical considerations
All of these methods may be applied to scans acquired in a clinical setting, so some comments are necessary regarding minimum sample sizes, scan times, and combination of data from different scanners (e.g., 1.5 versus 3 Tesla MRI). In large-scale research initiatives, such as the Alzheimer’s Disease Neuroimaging Initiative [98], and in clinical trials that use longitudinal neuroima-ging [12, 13, 22, 99], standardized protocols have been developed to ensure that data from multiple centers can be integrated reliably. To minimize burden on the participants, a total MRI scanning time of no longer than 30 minutes is typical, with no individual imaging sequence lasting more than 10 minutes [98]. For methods described in this paper, good quality 3D T1-weighted
MRIs are required, so priority is usually given to collecting at least one volumetric MRI scan with sufficient gray/white matter contrast and resolution (voxel dimensions , 1.5 mm along all axes) to resolve subtle degenerative brain changes. For the Alzheimer’s Disease Neuroimaging Initiative (ADNI) study, a some-what unusual step was taken of collecting two such 3D MP-RAGE scans, thereby maximizing the chance of collecting at least one high-quality dataset, and the best was chosen for analysis. Virtually all clinical enrolment sites for the ADNI study had access to scanners from one of three major vendors (GE Healthcare, Philips Medical Systems, or Siemens Medical Solutions), so a standard acquisition protocol was developed to accommodate most MRI platforms from each vendor. To ensure geometrical and intensity calibration across sites and across time, phantom-based monitoring of all scanners was used, in which a phantom with known geometry was scanned and measures had to remain within a prescribed tolerance over the course of the study. Several post-acquisition corrections were also used, such as 3D-distortion correction for geometrical warping due to
gradient non-linearity. Image intensity inhomogeneity correction was also found to be crucial in ensuring the stability of some longitudinal change measures [100, 101]. While the ADNI protocol may be considered a model for other multi-site trials, some open questions remain. First, it is not currently agreed whether 3 Tesla or 1.5 Tesla MRI scanning offers greater power to track degenerative brain changes, as the added contrast at higher fields is only exploited by some morphometric methods. Although higher fields offer higher contrast to noise for most signals of interest, some artifactual distortions are harder to control for at higher fields; these include geometric distortions at the temporal and frontal poles where artifacts are common. Second, we do not know the single best numeric index that can be derived from a scan to predict future cognitive decline, imminent conversion to dementia, or treatment response. When conventional volumetric measures of the hippo-campus are used, Jack et al. [81] estimated that in each arm of a therapeutic trial, only 21 subjects would be required to detect a 50% reduction in the rate of decline if hippocampal volumes were used as the outcome measure. This compared with 241 subjects if MMSE scores were used and 320 if the AD Assessment Scale Cognitive Subscale (ADAS-Cog) was used. There is significant interest in determining whether mapping methods (such as those proposed in this article) may provide numeric indices with greater power for tracking neurodegeneration when compared head-to-head with simpler measures, such as hippocampal volumes or overall rates of brain atrophy. Ongoing efforts are attempting to compute the minimal sample sizes required to distinguish AD, MCI, and controls, using statistical maps computed with TBM [102] and auto-mated hippocampal surface extraction in hundreds of subjects [103]. To compare the power of different methods that generate maps rather than numeric measures, approaches such as cumulative distribution plots and false-discovery rates can be useful [102]. Ultimately, which combination of scanner field strengths, analysis methods, and morphometric mea-sures gains widespread acceptance will depend on their statistical power, ease of use, and widespread avail-ability.
Conclusions
In this review, we surveyed a range of computational methods that can be used to monitor the progression of dementia, each focusing on specific features of the degenerative process. The approaches may be ranked in terms of their sensitivity to changes early in the disease process, the anatomical detail of the measures they provide, and their efficiency for use in large-scale epidemiological studies or clinical trials. Cortical thickness maps, for example, localize changes relative to gyral and sulcal landmarks and can be used to create time-lapse movies, but require some time investment in extracting and landmarking cortical surface models. If data collection is time-consuming and power requirements are high, this approach is still useful even in a clinical trial, as it may detect disease-related changes over intervals as short as 3 months [11]. Higher-throughput methods, such as TBM
and ventricular mapping, are well-suited to very large-scale studies. These may be the first mapping methods to be used to analyze routinely databases of images from thousands of subjects. As such, they are natural approaches for use in epidemiological studies of potential risk factors or protective factors, such as serum amyloid, cystatin C, or homocysteine levels, that may influence disease onset, conversion, or the normal aging process [104, 105]. Finally, newer imaging signals from amyloid-binding ligands, such as FDDNP and PIB, provide extraordinary promise for the early detection and tracking of Alzheimer’s disease. Unlike any MRI-based measure, these ligands detect brain changes that are linked with cognition even in asymptomatic subjects, presenting an opportunity to track therapeutic response years before dementia appears.
Acknowledgments
This work was supported by the National Institute on Aging (NIA), the National Library of Medicine, the National Institute for Biomedical Imaging and Bioengineering, the National Center for Research Resources, and the National Institute for Child Health and Development (AG016570, LM05639, EB01651, RR019771 and HD050735 to P.M.T.). L.G.A. was sup-ported by NIA K23 AG026803 (jointly sponsored by NIA, the American Federation for Aging Research, The John A. Hartford Foundation, the Atlantic Philanthropies, the Starr Foundation and an anonymous donor) and by NIA P50 AG16570.
References
1. Braak H, Braak E. Neuropathological stageing of Alzheimer-related changes. Acta Neuropathol (Berl) 1991;82:239–59.
2. Augustinack JC, van der Kouwe AJ, Blackwell ML, Salat DH, Wiggins CJ, Frosch MP, et al. Detection of entorhinal layer II using 7Tesla [corrected] magnetic resonance imaging. Ann Neurol 2005;57:489–94.
3. Mueller SG, Stables L, Du AT, Schuff N, Truran D, Cashdollar N, et al. Measurement of hippocampal subfields and age-related changes with high resolution MRI at 4T. Neurobiol Aging 2007;28:719–26.
4. Klunk WE, Engler H, Nordberg A, Wang Y, Blomqvist G, Holt DP, et al. Imaging brain amyloid in Alzheimer’s disease with Pittsburgh Compound-B. Ann Neurol 2004;55:306–19.
5. Small GW, Kepe V, Ercoli LM, Siddarth P, Bookheimer SY, Miller KJ, et al. PET of brain amyloid and tau in mild cognitive impairment. N Engl J Med 2006;355:2652–63. 6. Braskie MN, Klunder AD, Hayashi KM, Protas HD, Kepe V,
Miller KJ, et al. Dynamic trajectory of cortical plaque and tangle load correlates with cognitive impairment in normal aging and Alzheimer’s disease. J Neurosci 2007; submitted.
;
7. Mintun MA, Larossa GN, Sheline YI, Dence CS, Lee SY,Mach RH, et al. [11C]PIB in a nondemented population: potential antecedent marker of Alzheimer disease. Neurology 2006;67:446–52.
8. Mori S. Introduction to diffusion tensor imaging. Amsterdam: Elsevier; 2007.
9. Ashburner J, Csernansky JG, Davatzikos C, Fox NC, Frisoni GB, Thompson PM. Computer-assisted imaging to assess brain structure in healthy and diseased brains. Lancet Neurol 2003;2:79–88.
10. Apostolova LG, Thompson PM. Brain mapping as a tool to study neurodegeneration. Neurotherapeutics 2007;4: 387–400.
11. Thompson PM, Bartzokis G, Hayashi KM, Klunder AD, Lu PH, Edwards N, et al. Time-lapse mapping reveals different disease trajectories in schizophrenia depending on anti-psychotic treatment. J Neurosci 2007; submitted.
<
12. Fox NC, Black RS, Gilman S, Rossor MN, Griffith SG, Jenkins L, et al. Effects of Abeta immunization (AN1792) on MRI measures of cerebral volume in Alzheimer disease. Neurology 2005;64:1563–72.
13. Jack CR, Jr., Petersen RC, Grundman M, Jin S, Gamst A, Ward CP, et al. Longitudinal MRI findings from the vitamin E and donepezil treatment study for MCI. Neurobiol Aging 2007; Epub ahead of print: http://www.ncbi.nlm.nih.gov/ entrez/query.fcgi?cmd5Retrieve&db5PubMed&dopt5
Citation&list_uids517452062.
14. Barnes J, Scahill RI, Frost C, Schott JM, Rossor MN, Fox NC. Increased hippocampal atrophy rates in AD over 6 months using serial MR imaging. Neurobiol Aging 2007; Epub ahead of print: http://www.ncbi.nlm.nih.gov/entrez/ query.fcgi?cmd5Retrieve&db5PubMed&dopt5 Citation&-list_uids517368654.
15. Sowell ER, Peterson BS, Kan E, Woods RP, Yoshii J, Bansal R, et al. Sex differences in cortical thickness mapped in 176 healthy individuals between 7 and 87 years of age. Cereb Cortex 2007;17:1550–60.
16. Thompson PM, Hayashi KM, Sowell ER, Gogtay N, Giedd JN, Rapoport JL, et al. Mapping cortical change in Alzheimer’s disease, brain development, and schizophre-nia. Neuroimage 2004;23 Suppl 1:S2–S18.
17. Zeineh MM, Engel SA, Thompson PM, Bookheimer SY. Dynamics of the hippocampus during encoding and retrieval of face-name pairs. Science 2003;299:577–80. 18. Apostolova LG, Thompson PM, Rogers SA, Dinov ID,
Zoumalan CI, Steiner CA, et al. Surface feature-guided mapping of longitudinal FDG-PET changes in nondemen-ted elderly. Neurology 2008; submitnondemen-ted.
=
19. Tu Z, Zheng S, Yuille AL, Reiss AL, Dutton RA, Lee AD, et al. Automated extraction of the cortical sulci based on a supervised learning approach. IEEE Trans Med Imaging 2007;26:541–52.
20. Apostolova LG, Mosconi L, Thompson PM, Green AE, Mistur R, Tsui WH, et al. Subregional hippocampal atrophy predicts future decline to Alzheimer’s dementia in cogni-tively normal subjects. Neurology 2008; submitted.
>
21. Apostolova LG, Dutton RA, Dinov ID, Hayashi KM, Toga AW, Cummings JL, et al. Conversion of mild cognitive impairment to Alzheimer disease predicted by hippocam-pal atrophy maps. Arch Neurol 2006;63:693–9.
22. Apostolova LG, Thompson PM, Green AH, Jack CR, Harvey D, Petersen RC, et al. 3D analysis of hippocampal atrophy progression in mild cognitive impairment subjects. In: 132nd Annual Meeting of the American Neurological Association; 2007; Washington DC: American Neurological Association; 2007.
?
23. Gogtay N, Giedd JN, Lusk L, Hayashi KM, Greenstein D, Vaituzis AC, et al. Dynamic mapping of human cortical development during childhood through early adulthood. Proc Natl Acad Sci USA 2004;101:8174–9.
24. Duyckaerts C, Dickson DW. Neuropathology of Alzheimer’s disease. In: Dickson DW, editor. Neurodegeneration: the molecular pathology of dementia and movement disorders. Basel: ISN Neuropath Press; 2003: 47–65.
25. Apostolova LG, Cummings JL. The pathogenesis of Alzheimer’s disease: general overview. In: Uversky VN, Fink A, editors. Protein misfolding, aggregation and conformational diseases. New York: Springer Science+Business Media; 2007:3–29.
26. Ashburner J, Friston KJ. Unified segmentation. Neuroimage 2005;26:839–51.
27. Apostolova LG, Akopyan GG, Partiali N, Steiner CA, Dutton RA, Hayashi KM, et al. Structural correlates of apathy in Alzheimer’s disease. Dement Geriatr Cogn Disord 2007;24:91–7.
28. Frackowiak RSJ, Friston KJ, Frith C, Dolan R, Mazziotta JC, editors. Human Brain Function: Academic Press USA; 2007.
29. Friston KJ, Holmes A, Poline JB, Price CJ, Frith CD. Detecting activations in PET and fMRI: levels of inference and power. Neuroimage 1996;4:223–35.
30. Fischl B, Dale AM. Measuring the thickness of the human cerebral cortex from magnetic resonance images. Proc Natl Acad Sci USA 2000;97:11050–5.
31. Lerch JP, Pruessner JC, Zijdenbos A, Hampel H, Teipel SJ, Evans AC. Focal decline of cortical thickness in Alzheimer’s disease identified by computational neuroanatomy. Cereb Cortex 2005;15:995–1001.
32. Lohmann G, Preul C, Hund-Georgiadis M. Morphology-based cortical thickness estimation. Inf Process Med Imaging 2003;18:89–100.
33. Thompson PM, Mega MS, Woods RP, Zoumalan CI, Lindshield CJ, Blanton RE, et al. Cortical change in Alzheimer’s disease detected with a disease-specific popu-lation-based brain atlas. Cereb Cortex 2001;11:1–16. 34. Wright IC, McGuire PK, Poline JB, Travere JM, Murray RM,
Frith CD, et al. A voxel-based method for the statistical analysis of gray and white matter density applied to schizophrenia. Neuroimage 1995;2:244–52.
35. Narr KL, Bilder RM, Toga AW, Woods RP, Rex DE, Szeszko PR, et al. Mapping cortical thickness and gray matter concentration in first episode schizophrenia. Cereb Cortex 2005;15:708–19.
36. Ashburner J, Friston KJ. Voxel-based morphometry — the methods. Neuroimage 2000;11:805–21.
37. Good CD, Ashburner J, Frackowiak RS. Computational neuroanatomy: new perspectives for neuroradiology. Rev Neurol (Paris) 2001;157:797–806.
38. Bookstein FL. ‘‘Voxel-based morphometry’’ should not be used with imperfectly registered images. Neuroimage 2001;14:1454–62.
39. Thacker N. Tutorial: a critical analysis of VBM. Available at: http://www.tina-vision.net/docs/memos/2003-011.pdf. Last updated 08/28/05. Accessed 03/17/07.
40. Chiang MC, Dutton RA, Hayashi KM, Lopez OL, Aizenstein HJ, Toga AW, et al. 3D pattern of brain atrophy in HIV/AIDS visualized using tensor-based morphometry. Neuroimage 2007;34:44–60.
41. Ashburner J. A fast diffeomorphic image registration algorithm. Neuroimage 2007;38:95–113.
42. Wang YL, Chiang MC, Thompson PM. Automated surface matching using mutual information applied to Reimann surface structures. In: Duncan JS, Gerig G, editors. Medical image computing and computer-assisted intervention — MICCAI 2005, Lecture Notes in Computer Science. Berlin: Springer; 2005:666–74.
43. Shi Y, Tu Z, Reiss AL, Dutton RA, Lee AD, Galaburda AM, et al. Joint sulci detection using graphical models and boosted priors. Inf Process Med Imaging 2007;20:98–109. 44. Joshi AA, Shattuck DW, Thompson PM, Leahy RM.
Simultaneous surface and volumetric brain registration using harmonic mappings. IEEE Trans Med Imaging 2007;
in press.
@
45. Toga AW, Thompson PM. What is where and why is it important? (Peer reviewed invited commentary on a paper by PR Devlinj.) Neuroimage 2007; in press.
A
46. Apostolova LG, Steiner CA, Akopyan GG, Toga AW,mapping in mild cognitive impairment and mild Alzheimer’s disease. Arch Neurol 2007;64:1489–95. 47. Frisoni GB, Pievani M, Testa C, Sabattoli F, Bresciani L,
Bonetti M, et al. The topography of grey matter involve-ment in early and late onset Alzheimer’s disease. Brain 2007;130:720–30.
48. Ballmaier M, O’Brien JT, Burton EJ, Thompson PM, Rex DE, Narr KL, et al. Comparing gray matter loss profiles between dementia with Lewy bodies and Alzheimer’s disease using cortical pattern matching: diagnosis and gender effects. Neuroimage 2004;23:325–35.
49. Apostolova LG, Lu PH, Rogers S, Dutton RA, Hayashi KM, Toga AW, et al. 3D mapping of verbal memory perfor-mance in clinical and pre-clinical Alzheimer’s disease. 131st Annual Meeting of the American Neurological Association; Chicago, IL; 2006.
B
50. Apostolova LG, Lu PH, Rogers S, Dutton RA, Hayashi KM, Toga AW, et al. 3D mapping of language networks in clinical and pre-clinical Alzheimer’s disease. Brain Lang 2007;104:33–41.
51. Sowell ER, Peterson BS, Thompson PM, Welcome SE, Henkenius AL, Toga AW. Mapping cortical change across the human life span. Nature Neurosci 2003;6:309–15. 52. Salat DH, Buckner RL, Snyder AZ, Greve DN, Desikan RS,
Busa E, et al. Thinning of the cerebral cortex in aging. Cereb Cortex 2004;14:721–30.
53. Reisberg B, Franssen EH, Hasan SM, Monteiro I, Boksay I, Souren LE, et al. Retrogenesis: clinical, physiologic, and pathologic mechanisms in brain aging, Alzheimer’s and other dementing processes. Eur Arch Psychiatry Clin Neurosci 1999;249 Suppl 3:28–36.
54. Karas GB, Scheltens P, Rombouts SA, Visser PJ, van Schijndel RA, Fox NC, et al. Global and local gray matter loss in mild cognitive impairment and Alzheimer’s disease. Neuroimage 2004;23:708–16.
55. Apostolova LG, Lu PH, Rogers S, Dutton RA, Hayashi KM, Toga AW, et al. 3D mapping of mini-mental state examination performance in clinical and preclinical Alzheimer disease. Alzheimer Dis Assoc Disord 2006;20:224–31.
56. Thompson PM, Hayashi KM, de Zubicaray G, Janke AL, Rose SE, Semple J, et al. Dynamics of gray matter loss in Alzheimer’s disease. J Neuroscience 2003;23:994–1005. 57. Lu L, Leonard C, Thompson P, Kan E, Jolley J, Welcome S,
et al. Normal developmental changes in inferior frontal gray matter are associated with improvement in phonolo-gical processing: a longitudinal MRI analysis. Cereb Cortex 2007;17:1092–9.
58. Thompson PM, Cannon TD, Narr KL, van Erp T, Poutanen VP, Huttunen M, et al. Genetic influences on brain structure. Nat Neurosci 2001;4:1253–8.
59. Posthuma D, De Geus EJ, Baare WF, Hulshoff Pol HE, Kahn RS, Boomsma DI. The association between brain volume and intelligence is of genetic origin. Nat Neurosci 2002;5:83–4.
60. Haier RJ, Jung RE, Yeo RA, Head K, Alkire MT. Structural brain variation and general intelligence. Neuroimage 2004;23:425–33.
61. Kemp PM, Holmes C, Hoffmann SM, Bolt L, Holmes R, Rowden J, et al. Alzheimer’s disease: differences in technetium-99m HMPAO SPECT scan findings between early onset and late onset dementia. J Neurol Neurosurg Psychiatry 2003;74:715–9.
62. Vogt BA, Crino PB, Vogt LJ. Reorganization of cingulate cortex in Alzheimer’s disease: neuron loss, neuritic plaques, and muscarinic receptor binding. Cereb Cortex 1992;2:526–35.
63. Mosconi L, Pupi A, De Cristofaro MT, Fayyaz M, Sorbi S, Herholz K. Functional interactions of the entorhinal cortex:
an 18F-FDG PET study on normal aging and Alzheimer’s disease. J Nucl Med 2004;45:382–92.
64. Kril JJ, Halliday GM. Clinicopathological staging of frontotemporal dementia severity: correlation with regional atrophy. Dement Geriatr Cogn Disord 2004;17:311–5. 65. Rowe CC, Ng S, Ackermann U, Gong SJ, Pike K, Savage G,
et al. Imaging beta-amyloid burden in aging and dementia. Neurology 2007;68:1718–25.
66. Small GW, Protas HD, Huang SC, Kepe V, Siddarth P, Hayashi KM, et al. FDDNP binding values from cortical hemispheric surface maps correlate with MMSE scores. Alzheimer’s Association International Conference on the Prevention of Dementia. Washington, DC; 2007.
C
67. Lepore N, Brun CA, Chou YY, Chiang MC, Dutton RA,Hayashi KM, et al. Generalized tensor-based morphometry of HIV/AIDS using multivariate statistics on strain matrices and their application to HIV/AIDS. IEEE Trans
Med Imaging 2007;in press.
EX
68. Fox NC, Cousens S, Scahill R, Harvey RJ, Rossor MN. Using serial registered brain magnetic resonance imaging to measure disease progression in Alzheimer disease: power calculations and estimates of sample size to detect treatment effects. Arch Neurol 2000;57:339–44.
69. Thompson PM, Woods RP, Mega MS, Toga AW. Mathematical/computational challenges in creating deformable and probabilistic atlases of the human brain. Hum Brain Mapp 2000;9:81–92.
70. Brambati SM, Renda NC, Rankin KP, Rosen HJ, Seeley WW, Ashburner J, et al. A tensor based morphometry study of longitudinal gray matter contraction in FTD. Neuroimage 2007;35:998–1003.
71. Thompson PM, Dutton RA, Hayashi KM, Toga AW, Lopez OL, Aizenstein HJ, et al. Thinning of the cerebral cortex visualized in HIV/AIDS reflects CD4+ T lymphocyte decline. Proc Natl Acad Sci USA 2005;102:15647–52. 72. Chiang MC, Leow AD, Dutton RA, Barysheva M, Rose SE,
McMahon KL, et al. Fluid Registration of diffusion tensor images using information theory. IEEE Trans Med Imaging
2007; in press.
EO
73. Leow AD, Chiang MC, Yanovsky I, Lee AD, Lu A, Klunder AD, et al. Statistical properties of Jacobian maps and inverse-consistent deformations in nonlinear image regis-tration. IEEE Trans Med Imaging 2007;26:822–32.
74. Lepore N, Brun CA, Pennec X, Chou YY, Lopez OL, Aizenstein HJ, et al. Mean template for tensor-based morphometry using deformation tensors. Ayache N, Ourselin S, Maeder AJ, editors. Medical image computing and computer-assisted intervention — MICCAI 2007, Part II. Lecture Notes in Computer Science. Berlin: Springer 2007; 826–33.
75. Fillard P, Arsigny V, Pennec X, Hayashi KM, Thompson PM, Ayache N. Measuring brain variability by extrapolat-ing sparse tensor fields measured on sulcal lines. Neuroimage 2007;34:639–50.
76. Brun CA, Lepore N, Pennec X, Chou YY, Lopez OL, Aizenstein HJ, et al. Comparison of standard and Reimannian elasticity for tensor-based morphometry in HIV/AIDS. In: Ayache N, Ourselin S, Maeder AJ, editors. Medical image computing and computer-assisted interven-tion — MICCAI 2007. Lecture Notes in Computer Science. Berlin: Springer 2007.
77. Hua X, Leow A, Lee SE, Klunder AD, Jack CR, Toga AW, et al. 3D characterization of brain atrophy in Alzheimer’s disease and mild cognitive impairment using tensor-based morphometry. Neuroimage 2007;submitted.
EP
78. Jack CR, Jr., Petersen RC, Xu Y, O’Brien PC, Smith GE, IvnikRJ, et al. Rates of hippocampal atrophy correlate with change in clinical status in aging and AD. Neurology 2000;55:484–89.
79. Jack CR, Jr., Petersen RC, Xu Y, O’Brien PC, Smith GE, Ivnik RJ, et al. Rate of medial temporal lobe atrophy in typical aging and Alzheimer’s disease. Neurology 1998;51:993–9. 80. Du AT, Schuff N, Kramer JH, Ganzer S, Zhu XP, Jagust WJ,
et al. Higher atrophy rate of entorhinal cortex than hippocampus in AD. Neurology 2004;62:422–7.
81. Jack CR, Jr., Shiung MM, Gunter JL, O’Brien PC, Weigand SD, Knopman DS, et al. Comparison of different MRI brain atrophy rate measures with clinical disease progression in AD. Neurology 2004;62:591–600.
82. Thompson PM, Hayashi KM, De Zubicaray GI, Janke AL, Rose SE, Semple J, et al. Mapping hippocampal and ventricular change in Alzheimer disease. Neuroimage 2004;22:1754–66.
83. Becker JT, Davis SW, Hayashi KM, Meltzer CC, Toga AW, Lopez OL, et al. Three-dimensional patterns of hippocam-pal atrophy in mild cognitive impairment. Arch Neurol 2006;63:97–101.
84. Frisoni GB, Sabattoli F, Lee AD, Dutton RA, Toga AW, Thompson PM.In vivoneuropathology of the hippocampal formation in AD: a radial mapping MR-based study. Neuroimage 2006;32:104–10.
85. Ballmaier M, Narr KL, Toga AW, Elderkin-Thompson V, Thompson PM, Hamilton L, et al. Hippocampal morphol-ogy distinguishes late-onset from early-onset elderly depression. Am J Psychiatry 2007;165:229–37.
86. Sabattoli F, Boccardi M, Lee AD, Dutton RA, Thompson PM, Frisoni GB. Hippocampal shape changes in dementia with Lewy bodies. Neuroimage 2007; submitted.
EQ
87. Scher AI, Xu Y, Korf ESC, White LR, Scheltens P, Toga AW, et al. Hippocampal morphometry in Alzheimer’s vs. vascular dementia: a population-based study. Neuroimage 2007; submitted.
ER
88. Csernansky JG, Joshi S, Wang L, Haller JW, Gado M, Miller JP, et al. Hippocampal morphometry in schizophrenia by high dimensional brain mapping. Proc Natl Acad Sci U S A 1998;95:11406–11.
89. Csernansky JG, Wang L, Joshi S, Miller JP, Gado M, Kido D, et al. Early DAT is distinguished from aging by high-dimensional mapping of the hippocampus. Dementia of the Alzheimer type. Neurology 2000;55:1636–43.
90. Wang L, Joshi SC, Miller MI, Csernansky JG. Statistical analysis of hippocampal asymmetry in schizophrenia. Neuroimage 2001;14:531–45.
91. Fischl B, Salat DH, van der Kouwe AJ, Makris N, Segonne F, Quinn BT, et al. Sequence-independent segmentation of magnetic resonance images. Neuroimage 2004;23 Suppl 1:S69–S84.
92. Morra JH, Tu Z, Apostolova LG, Green A, Toga AW, Thompson PM. Comparison of Adaboost and support vector machines for detecting Alzheimer’s disease through automated hippocampal segmentation. IEEE Trans Med Imaging 2007;submitted.
ES
93. Apostolova LG, Lu P, Rogers S, Dutton RA, Hayashi KM, Toga AW, et al. 3D mapping of language networks in clinical and pre-clinical Alzheimer’s disease. Brain Lang 2007;104:33–41.
94. Carmichael OT, Kuller LH, Lopez OL, Thompson PM, Dutton RA, Lu A, et al. Ventricular volume and dementia progression in the Cardiovascular Health Study. Neurobiol Aging 2007;28:389–97.
95. Carmichael OT, Thompson PM, Dutton RA, Lu A, Lee SH, Lee JY, et al. Mapping ventricular changes related to dementia and mild cognitive impairment in a large community-based cohort. In: Proceedings of the 3rd IEEE International Symposium on Biomedical Imaging: Macro to
Nano; 2006.
ET
96. Chou Y, Lepore N, de Zubiricaray GI, Carmichael OT, Becker J, Toga AW, et al. Automated ventricular mapping with multi-atlas fluid image alignment reveals genetic effects in Alzheimer’s disease. Neuroimage 2007;[Epub ahead of print] DOI:10.1016/j.neuroimage.2007.11.047 97. Chou Y, Lepore N, Zubiricay GI, Rose SE, Carmichael OT,
Becker JT, et al. Automated 3D mapping and shape analysis of the lateral ventricles via fluid registration of multiple surface-based atlases. Proceedings of the 2007 IEEE International Symposium on Biomedical Imaging: From Nano to Macro;in press.
98. Jack CR, Jr., Bernstein MA, Fox NC, Thompson PM, Alexander GE, Harvey D, et al. The Alzheimer’s Disease Neuroimaging Initiative (ADNI): the MR Imaging protocol. J Magn Reson Imaging 2007; in press.
EU
99. Lieberman JA, Tollefson GD, Charles C, Zipursky R,Sharma T, Kahn RS, et al. Antipsychotic drug effects on brain morphology in first-episode psychosis. Arch Gen Psychiatry 2005;62:361–70.
100. Boyes RG, Gunter JL, Frost C, Janke AL, Yetman T, Hill DLG, et al. Quantitative analyses of N3 on 3Tesla scanners with multichannel phased array coils. Neuroimage 2007;
in press.
EV
101. Leow AD, Klunder AD, Jack CR, Jr., Toga AW, Dale AM, Bernstein MA, et al. Longitudinal stability of MRI for mapping brain change using tensor-based morphometry. Neuroimage 2006;31:627–40.
102. Hua X, Leow A, Lee SE, Klunder AD, Jack CR, Toga AW, et al. 3D characterization of brain atrophy in Alzheimer’s disease and mild cognitive impairment using tensor-based morphometry. Neuroimage 2008;submitted.
EW
103. Morra JH, Tu Z, Apostolova LG, Green AE, Avedissian C,Madsen SK, et al. Automated 3D mapping of hippocampal atrophy and its clinical correlates in 400 subjects with Alzheimer’s disease, mild cognitive impairment and elderly controls. Neuroimage 2008; submitted.
FX
104. Carmichael OT, Kuller LH, Lopez OL, Thompson PM,Dutton RA, Lu A, et al. Cerebral ventricular changes associated with transitions between normal cognitive function, mild cognitive impairment, and dementia. Alzheimer Dis Assoc Disord 2007;21:14–24.
105. Lopez OL, Kuller LH, Becker JT, Dulberg C, Sweet RA, Gach HM, et al. Incidence of dementia in mild cognitive impairment in the cardiovascular health study cognition study. Arch Neurol 2007;64:416–20.
The British Journal of Radiology
36 Portland Place
London W1B 1AT, UK
Tel.: +44 (0)207 307 1400
Fax: +44 (0)207 307 1414
E-mail: [email protected]
Dear Author,Please find enclosed a proof of your article ‘‘Computational anatomic methods as applied to aging and
dementia’’for checking.
When reading through your proof, please check carefully authors’ names, scientific data, data in tables, any mathematics and the accuracy of references. Please do not make any unnecessary changes at this stage. All necessary corrections should be marked on the proof at the place where the correction is to be made; please write the correction clearly in the margin (if in the text they may be overlooked).
Please also check the quality of the figures. For information about electronic image preparation, see the instructions for authors (http://bjr.birjournals.org/misc/ifora.pdf).
Any queries that have arisen during preparation of your paper for publication are listed below and indicated on the proof. Please provide your answers when returning your proof.
Please return your proof by fax (+44 (0)207 307 1414) or post within 3 days of receipt.
Query no. Query
1 [Reference?]
1 AuQ1 What is the updated status of this paper?
2 AuQ2 What is the updated status of this paper?
3 AuQ3 What is the updated status of this paper?
4 AuQ4 What is the updated status of this paper?
5 AuQ5 Would it be better to replace this by the paper of the same title in Brain?
6 AuQ6 What is the updated status of this paper?
7 AuQ7 What is the updated status of this article?
8 AuQ8 Is this in a published proceedings?
9 AuQ9 Is this in a published proceedings?
10 AuQ10 What is the updated status of this paper?
12 AuQ12 What is the updated status of this paper?
13 AuQ13 What is the updated status of this paper?
14 AuQ14 What is the updated status of this paper?
15 AuQ15 What is the updated status of this paper?
16 AuQ16 Please give editors if any, place of publication, publisher and page range.
17 AuQ17 What is the updated status of this paper?
18 AuQ18 What is the updated status of this paper?
19 AuQ19 What is the updated status of this paper?
20 AuQ20 What is the updated status of this paper?
21 AuQ21 [missing from reference list]
22 AuQ22 [missing from reference list]
23 AuQ23 [missing from reference list]
24 AuQ24for each figure legend, please indicate the number of the reference that you
citing, rather than giving the name of the authors. Where these have not been included into the reference list, please provide the full citation so that these references can be added to the list
BJR Advance
declaration
Your manuscript may be published ahead of print in this proof form online on BJR Advance (for further
information contact the BJR editorial office or visit http://bjr.birjournals.org/pap.dtl). If you, the
corresponding author, do not want this manuscript, ‘‘Computational anatomic methods as applied to
aging and dementia’’published in proof form online atBJR Advance, please sign below.
Signature
