|Home | About | Journals | Submit | Contact Us | Français|
Alzheimer’s disease (AD) is the most common neurodegenerative disorder for those 65 years or older, currently affects 4.5 million Americans, and is predicted to rise to 13.2 million by the year 2050 in the U.S. alone. Neuroimaging and brain mapping techniques provide extraordinary power to understand AD, providing spatially detailed information on the extent and trajectory of the disease as it spreads in the living brain. Computational anatomy techniques, applied to large databases of brain MRI scans, reveal the dynamic sequence of cortical and hippocampal changes with disease progression, and how they relate to cognitive decline and future clinical outcomes. Those who are mildly cognitively impaired, in particular, are at a five-fold increased risk of imminent conversion to dementia, and show specific structural brain changes that are predictive of imminent disease onset. We review the principles and key findings of several new methods for assessing brain degeneration, including voxel-based morphometry, tensor-based morphometry, cortical thickness mapping, hippocampal atrophy mapping and automated methods for mapping ventricular anatomy. Applications to AD and other dementias are discussed, with a brief review of related findings in other neurological and neuropsychiatric illnesses, including epilepsy, HIV/AIDS, schizophrenia and disorders of brain development.
The desire to create maps that localize cognitive functions and disease-related changes to discrete brain regions has long fascinated the scientific world. Perhaps the most celebrated brain mapping achievement was Brodmann’s cytoarchitectonic map that distinguished 52 cortical subregions based on their differences in thickness, lamination, neuronal type and staining characteristics.1 Since the advent of non-invasive neuroimaging, imaging brain maps have evolved to consist of sophisticated multi-modal and multi-dimensional digital atlases that cover the human life span and represent a variety of diseases and their progression.2
Neurodegenerative diseases are a large group of neurologic disorders that share a similar pathophysiology. The vast majority are associated with the intra- or extracellular deposits of abnormal proteins and present with significant neurologic and cognitive dysfunction.3 This review will mainly focus on the most prevalent neurodegenerative disorders — Alzheimer’s disease (AD) and other related dementias. We will also briefly summarize recent brain mapping advances in some other progressive neurologic and psychiatric disorders such as schizophrenia, epilepsy, HIV/AIDS, and abnormal brain development.
AD is the most common neurodegenerative disorder in those 65 years or older. It results from the pathological accumulation of two highly amyloidogenic proteins, A-beta and tau.3 It currently affects 4.5 million Americans, and is projected to affect 13.2 million in year 2050 in the U.S. alone.4 Clinically, AD is preceded by a transitional cognitive state called mild cognitive impairment (MCI) in which persons experience memory and/or other cognitive changes but continue to lead an independent lifestyle.5 As AD pathology builds up, cognitive decline relentlessly progresses. Autonomy in daily living becomes increasingly difficult, and most patients become fully dependent upon others for even the most basic activities of daily living such as toileting or eating. The profound social, economic and emotional burden of AD has brought AD research to the forefront as one of the most pressing health problems of the 21st century. Currently available treatment options for AD patients are limited to symptomatic therapeutic agents that have not convincingly been shown to slow disease progression. With several disease modifying agents at various stages of clinical testing, researchers are working in parallel to develop the most sensitive combinations of clinical and biologic markers for early or even pre-clinical diagnosis, and to track disease progression and neurotherapeutic effectiveness.
The earliest brain changes in AD occur in the entorhinal cortex and the hippocampus where neurofibrillary tangles (NFTs) initially accumulate. These early events are followed by a spread of NFTs to the inferior and lateral temporal, then parietal, occipital and finally frontal cortices. The amyloid neuritic plaques originate in the lateral temporal neocortex and then spread to the inferior temporal, parietal, occipital and frontal association cortices6,7 (Fig. 1, top).
As the earliest sites to accumulate NFT pathology, the hippocampus and entorhinal cortex have become the most studied brain regions in AD. They have been traditionally examined in structural brain images with a region-of-interest technique (ROI). ROI techniques typically compute an overall volume for each brain structure, based on manual or automated delineations in serial sections of a subject’s MRI. To ensure adequate reliability in defining the structure boundaries, a formalized anatomical protocol is typically used, with detailed rules to guide image analysts in segmenting the structure with high reliability and consistency across raters. Automated, computer vision algorithms have also been developed to delineate these structures automatically on MRI, but none of these are currently widely used.8-10 Reliable automated segmentations are difficult to obtain in the medial temporal lobe, as anatomical boundaries are complex and image contrast is often low and is even affected by the disease process. The a priori knowledge that these structures are involved in MCI and AD makes ROI-based analysis a powerful approach, but it does not permit detailed investigation of the underlying complex structure of the hippocampus. Nevertheless it was through this technique that the hippocampus and entorhinal cortex were established as the most prominent imaging biomarkers in AD.
Hippocampal atrophy occurs in normal aging at an estimated rate of around 1.6-1.7% annually.11,12 In MCI and AD the yearly atrophy rates are several-fold higher (2.8% for stable MCI, 3.7% for MCI who convert to AD and 3.5-4% for AD subjects).11,12 Additionally, smaller average hippocampal volume in MCI carries an increased risk (odds ratio of 1.75) for future conversion to AD over a 1.2 to 4.8 years follow-up period.13 Similarly, in genetically predisposed AD patients with autosomal dominant mutations, progressive hippocampal atrophy is detected as much as 5.5 years before clinical diagnosis of AD can be established.14 However, as useful a simple measure as regional volume can be, it does not adequately capture the complex profile of disease progression which involves selective changes in certain hippocampal subfields. From the entorhinal cortex, NFTs spread first to the CA1 and subiculum, then to the CA2/3 and finally to the CA4 hippocampal subfield.15 Newer techniques (see below section, Brain mapping as a tool to study neurodegeneration) can now better localize discrete subregional changes that can predict disease progression in MCI. Subregion-specific changes between AD and MCI can also be related to known cytoarchitectural subdivisions.16,17
The entorhinal cortex has likewise been extensively studied with ROI techniques and its predictive value for future cognitive decline in MCI is well established.18-21 The annual rate of 1.4% entorhinal cortical atrophy in normal aging is far surpassed by the pathological 7% average annual loss observed in AD.22 The ROI technique is well suited for the simple sheet-like structure of the entorhinal cortex. However, significant technical challenges stem from its substantial anatomical variability, ambiguous boundaries and the lack of a widely embraced tracing protocol. The latter makes between-study comparisons challenging. A major limitation of the ROI approach for the entorhinal cortex lies in its focus on one important but small area of the brain in a neurodegenerative disease that affects the cortex in its entirety.
Voxel-based morphometry (VBM) is a newer image analysis technique that can simultaneously identify multiple areas of cortical and subcortical degeneration. It has provided significant insights into the gray matter changes in AD and more recently in MCI. Initially applied to studies of schizophrenia,23 VBM was subsequently implemented in the widely-used Statistical Parametric Mapping software package.24-26 The standard VBM method classifies each subject’s 3D brain MRI scan into individual maps that are representative of gray and white matter and CSF tissue classes, and all subjects’ gray matter images are then aligned to a common template before averaging the results. The intensity of these spatially normalized gray matter maps is spatially smoothed with a filter,24,27,28 and group differences or associations with cognitive scores are assessed by fitting a statistical model (typically multiple regression) at each image location. Statistical differences are then shown as a spatial map. The most common statistical analysis in VBM fits the general linear model (GLM) to the data (gray matter density) from all subjects at each image location, or voxel. This identifies voxels where tissue density relates to covariates of interest such as diagnosis, cognitive scores, etc., after discounting confounding effects (such as age, sex, educational level, etc.). As in more traditional volumetric studies, a measure of total gray matter or overall cerebral volume may be used as a covariate of interest, to detrend brain size or global atrophy effects from the data. Including the overall amount of gray matter as a covariate in the model enables the detection of gray matter that are regionally specific, beyond any global differences.
Some caveats are required in interpreting the findings of VBM because it can infer local anatomical differences (i) from systematic image registration errors in one group relative to the other; and (ii) from systematic shifts in unaffected regions that result from differences in truly affected structures.29,30 In response to these criticisms, the authors improved the method to better reflect volumetric differences between subjects, and to avoid incorrect localization of group differences that can occur due to imperfect registration of the images.25 With the original VBM approach, comparisons of gray matter density alone across subjects do not fully reflect volume differences between subjects, as they do not contain information on the expansions and contractions needed to match the subject to the common template. The “modulated”, or optimized, VBM method27 addresses this issue by multiplying (modulating) the voxel intensity values of the spatially normalized gray matter maps by the corresponding Jacobian determinant (expansion factor) of the deformation fields, so that the total amount of gray matter at each voxel, which is the product of the gray matter density and the volume of the deformed voxel, remains unchanged during warping. The method then reveals systematic volume differences for any brain region, assuming the warping process matches anatomy correctly across subjects. A conceptually similar approach (RAVENS) has been proposed by Davatzikos et al.31 and has been used successfully in studies of degenerative disease.
VBM has some limitations, stemming from the inherently low spatial resolution caused by spatial smoothing of the gray matter maps in order to control for inter-individual cortical variability. Governed by the full width at half maximum of the smoothing kernel (most commonly 12 mm; see Salmond et al., 2002,32 for different choices), the spatial resolution of VBM is limited. By the ‘matched filter theorem’, larger smoothing filters make it easier to detect diffuse or widespread effects, at the expense of blurring observations from different anatomic regions. This smoothing (1) partly accounts for registration errors and reduces inter-individual variance, increasing detection sensitivity; and (2) makes the data a better approximation to a Gaussian random field. This normality of the residuals is a requirement if parametric statistics are used in later processing.
Several VBM studies have documented gray matter atrophy in the temporal, posterior cingulate and precuneal cortex in AD relative to normal controls.33-36 As expected, between MCI and AD and MCI and normal aging these differences are less profound. Thus in agreement with clinical observations which place MCI between normal cognitive aging and AD based on atrophy severity, MCI likewise appears as intermediate between normal aging and AD.37-39 Progressive brain atrophy has been documented in MCI subjects who have declined (and even in those who have remained stable), providing evidence for a neurodegenerative etiology and advancing disease pathology in the latter as well.40 Whole brain atrophy and/or ventricular size are powerful predictors of future cognitive decline throughout the spectrum of normal aging to onset of AD.13 The predictive role of VBM was persuasively demonstrated in a study of 28 asymptomatic carriers of dominantly inherited genetic forms of early-onset AD. Subjects who converted to dementia had an annual brain volume loss of 5% compared to only 0.1% in those who remained cognitively intact.41 Ventricular enlargement appears to have a higher predictive value in the earlier transition stages as opposed to the later stage – that is, from normal cognition to MCI or AD and less so from MCI to AD.13,42 VBM has also been used to reveal the spatial profile of aging and gender effects in normal subjects,28,43,44 frontotemporal and Lewy body dementia,45-48 Parkinson’s disease, and even herpes simplex encephalitis.49
For analysis of cortical gray matter atrophy, VBM may not be optimal. Gyral and sulcal features are highly convoluted and individual gyri or sulci are not readily distinguished based on image intensity alone. As a result, most automated image warping algorithms — which rely on image intensities to match anatomical features — fail to match cortical gyri across subjects when making morphometric comparisons. This leads to a lack of power in detecting and localizing subtle cortical differences. In response, surface-based modeling methods have used specific cortical features as constraints to guide the matching of anatomy across subjects. Some of these methods also compute the cortical gray matter thickness directly in 3D based on an explicit model of the cortical surface geometry or on successive coding of voxels within the gray matter sheet.50,51
Several powerful brain mapping techniques have emerged over the last decade. Many rely on computational anatomy – a mathematical brain modeling approach where brain surfaces and subvolumes are viewed as complex geometrical patterns and are modeled as 3D continuous mesh models, or deformable shapes, that can be averaged and combined across subjects, and on which statistics can be defined (see Gee and Thompson, 2007, for a recent review of the field).52 The anatomical images thus represent deformable templates that are elastically or fluidly transformed into a similar shape (most commonly onto the study group average, an atlas average or another subject’s brain shape). Some of these techniques explicitly model brain anatomy – for instance, they may utilize surface landmarks as constraints (e.g., sulci). These techniques readily allow for an accurate alignment of surface specific geometrical patterns (e.g., gyri) and help precisely co-localize identical cortical and subcortical regions without sacrificing the underlying details contained in the measure of interest (cortical thickness, gray matter density, functional activation, etc.; Figs. 1 and and2).2). The resulting anatomical co-registration increases the power to characterize cortical and subcortical disease patterns and detect subtle disease-associated changes.
One newer technique to localize changes within the brain is known as tensor-based morphometry (TBM). TBM relies on a fully automated fluid warping to spatially register cortical and subcortical structures in cross-sectional and longitudinal imaging datasets (see Chiang et al., 2007, for mathematical details).53 When TBM is used to map brain changes over time, the follow-up (repeat) image is first globally aligned to the baseline scan, and then a 3D elastic or fluid image deformation is used to maximize the mutual information (or a related information-theoretic measure of correspondence) between the two consecutive scans. This fully 3D deformation reconfigures the baseline anatomy into the shape of the follow-up scan. The expansion or contraction at each image voxel is computed from the deformation field (using the Jacobian of the deformation field to produce a ‘voxel compression map’ or ‘tensor map’). In this map, contraction implies atrophy; expansion implies local growth or dilation.54 A color map then displays these changes on the follow-up scan (Fig. 3). Brain changes can be mapped in patients scanned longitudinally, such as individuals with semantic dementia,55 or normal subjects scanned over short intervals to assess longitudinal stability of a scanning protocol.55 TBM has high throughput and sensitivity, making it attractive for gauging brain changes in large population studies and clinical trials. TBM may also be used for cross-sectional studies; in that case, many individual images are fluidly transformed to match a common brain template, and the applied expansions and contractions are analyzed to identify systematic volume and shape differences in one subject group versus another. Used cross-sectionally, TBM has provided powerful visualizations of brain deficits, and their clinical correlates, in populations with HIV/AIDS (Fig. 7),53 Fragile X syndrome,56 and Williams syndrome.57
New statistical methods are also emerging to make TBM more sensitive to subtle neurodegenerative changes. If only the compression or expansions are analyzed, most of the information on atrophy, which may be directionally-dependent (i.e., non-isotropic), is discarded. So-called “Lie group” methods can detect neurodegeneration with greater power, e.g. in studies of HIV/AIDS,58 as they draw upon the full multi-dimensional information available in the deformation tensors, and model the tensor statistics as a statistical process on a Riemannian manifold (an approach called ‘generalized TBM’). Thus TBM can detect and visualize focal areas of structural gray and white matter and subcortical nuclear changes.53,55,58
3D modeling techniques have enabled the localization of disease-associated changes at the subfield level of the hippocampus. Some of these techniques have relied on radial measurements that assess the thickness of a structure (Fig. 4),59 others on large-deformation high-dimensional techniques60 or fluid-based registration of structural data within or across subjects.42,61 Rather than building 3D average surface meshes from manually-derived traces of the hippocampus, the high-dimensional modeling approach uses an elastic registration method to deform a hippocampal surface template onto new subjects’ scans, adapting to anatomical shape differences. As such, it is a variant of the TBM approach, which uses fluid registration to measure volume change, and the hippocampal radial atrophy technique, which uses the surface models to measure local changes on the hippocampus. Aiming to detect substructural changes within the hippocampus, our group reported a technique for unfolding of the hippocampal gray matter sheet with subsequent alignment of the hippocampal subfields.62 Another research group delineated the hippocampal subfields on T2-weighted MRI data obtained at 4T on three hippocampal slices and studied the effects of aging on the derived partial subfield volumes.63 Ultimately, the best approach to delineate hippocampal subfields in vivo may require higher-field MRI – at field strengths as high as 7T – in which the layers of the entorhinal cortex can be visualized.64
Neuropathological studies describe hippocampal NFT pathology as a stage-like process affecting the subiculum and CA1 subfields first, followed by the CA2/3 and finally the CA4 subfields.15 In 2004, Thompson et al.59 applied a hippocampal ‘radial atrophy mapping’ technique to AD and normal elderly subjects and showed profound differences between the two groups. Using the same technique, Frisoni et al. demonstrated the CA1 area and parts of the subiculum to show 15-20% atrophy in AD relative to normal controls.65 More recently, Apostolova et al.17 showed that MCI individuals with more severe CA1 and subiculum involvement are more likely to convert to AD (Fig. 5). Atrophy in the subiculum and CA1 area in this study also showed strong correlation with impaired delayed recall on the California Verbal Learning Test.17 In a follow-up study, the same group compared the hippocampal data of MCI and AD subjects. In agreement with neuropathological data, the hippocampal formation in AD was found to be more involved - with greater atrophy of the CA1 and additional atrophy of the CA2 and CA3 subfields.16 Becker et al.66 also used the hippocampal radial atrophy technique in a cross-sectional study of single domain amnestic MCI, multiple domain amnestic/nonamanestic MCI, AD and normal elderly subjects. The AD and single domain amnestic MCI groups showed significant subicular atrophy relative to normal controls. The AD group had also involvement of the CA1 subfield. In contrast, the multiple domain MCI group did not differ from the normal control group in either volumes or 3D hippocampal maps.66 Similar findings have been reported by researchers using the related large-deformation high-dimensional modeling approach. With this technique, two studies compared mild AD subjects with a Clinical Dementia Rating Scale of 0.5 to age-matched controls. They demonstrated predominant lateral edge hippocampal atrophy corresponding to the CA1 subfield.60,67
Apolipoprotein E4 (ApoE4) is the most prominent genetic risk factor for sporadic AD. A longitudinal study of cognitively normal elderly who were genotyped demonstrated that ApoE4 carriers had higher hippocampal atrophy rates relative to non-carriers.68 In a cortical thickness study, Burggren et al.69 reported that cognitively normal ApoE4 carriers have significantly thinner entorhinal cortex and focal hippocampal atrophy limited to the subiculum, relative to cognitively normal non-carriers.
The cortical pattern matching technique70 is an advanced computational anatomy imaging methodology that uses sulcal constraints while elastically deforming each subject’s cortical morphology into the group average representation. Cortical surface models, derived from structural MRI data, may be used to display a wide variety of imaging data such as cortical thickness, gray matter density, functional activation data, or metabolic data from PET scans. In the first manuscript published using this technique, Thompson and colleagues compared 12 AD (Mini-Mental State Examination (MMSE) score at baseline=18, MMSE at 1.5 year follow-up=14) and 14 age-matched cognitively intact subjects, scanned twice 1.5 years apart.70 The baseline gray matter density comparison maps demonstrated significant (> 15 %) atrophy in the lateral temporal, parietal and parieto-occipital cortices of AD patients. At follow-up, in agreement with the well-established disease progression sequence, these changes were shown to engulf the frontal lobes. On the mesial surface profound atrophy was noted for the left hemisphere at baseline while the right was relatively preserved with more focal precuneal and mesial temporal atrophy at baseline that spread to the cingulate cortex at follow-up. A time-lapse movie of the dynamics of AD spreading through the neocortex can be found at http://www.loni.ucla.edu/~thompson/AD_4D/dynamic.html.
Computational anatomy techniques may offer greatly improved sensitivity for detection of disease-induced group differences. A recent study compared amnestic MCI and mild AD subjects (Fig. 1).71 Despite the small cognitive differences between the two groups (mean MMSE difference of 4.4), the authors reported highly significant greater cortical atrophy in mild AD versus MCI. Most affected were the entorhinal, inferior and lateral temporal and the medial parietal/posterior cingulate cortices, followed by the lateral parietal, occipital and finally the frontal association cortices. This pattern concurs exceptionally well with the neuropathologic evidence for cortical disease progression (Fig. 1).6,7 Using an alternative cortical thickness mapping approach, Lerch et al.72 compared 19 mild to moderate AD subjects (mean MMSE=21.2, range 10-29) to 17 healthy volunteers. The AD patients had globally thinner cortical mantles (3.1 ± 0.28 mm versus 3.74 ± 0.32 mm in the controls). Statistically, the most profound differences were seen in the inferior and lateral temporal, precuneus, posterior cingulate and temporo-occipital association cortices.72 The reported regional effects were similar to the ones reported by Apostolova et al.71 in the study comparing mild AD (mean MMSE=23.8) to MCI patients (mean MMSE=28.2). A larger follow-up Freesurfer study73 including 34 healthy controls, 62 MCI and 42 AD patients revealed that, as predicted, MCI patients have intermediate cortical thickness values relative to normal controls and AD subjects (normal controls versus MCI mean cortical thickness difference = 0.18 mm; MCI versus AD mean cortical thickness difference = 0.26 mm). When comparing MCI versus normal controls, strongest effect sizes were seen for the entorhinal and lateral occipito-temporal gyri. In the AD versus normal controls comparison, significant differences were seen throughout the lateral surface of the brain with relative sparing of the primary sensorimotor and occipital regions. Differences for the mesial hemispheric surface were not reported.73
A very recent cortical pattern matching study74 examined the difference in severity and localization of cortical atrophy in subjects with sporadic early-onset AD (EOAD; <65 years of age) versus late-onset AD (LOAD; >65 years of age). EOAD subjects showed widespread atrophic changes relative to cognitively intact age-matched controls, but LOAD subjects showed a less severe and more focal pattern of entorhinal, parahippocampal, inferior temporal, posterior cingulate/precuneal, and lateral temporal changes. These findings imply that age of onset has a powerful influence on the severity of disease-induced changes required to produce cognitive symptoms. Younger AD subjects displayed higher tolerance to pathological burden, suggesting that they may have higher cognitive reserve.
Cortical neurodegeneration correlates with cognitive decline. Our group has so far reported correlations for one global measure of cognitive function – the MMSE,75,76 a verbal fluency and a picture naming test (animal fluency and Boston Naming test),77 as well as delayed recall of verbal information (California Verbal Learning Test).78 Strong associations between MMSE and cortical atrophy were seen in the entorhinal, parahippocampal, precuneus, superior parietal and subgenual cingulate/orbitofrontal cortices.70,75 Impaired picture naming and semantic fluency correlated with atrophy in perisylvian cortical areas thought to house lexical phonologic and semantic representations and process the receptive and articulatory aspects of language.77 Delayed recall performance was associated with bilateral precuneal, entorhinal, parahippocampal, inferior temporal as well as the left temporo-occipital cortical atrophy. These areas have been implicated in retrieval of verbal information and storing of lexical phonologic representations.78
Some ROI-based and VBM studies have suggested that ventricular volumes and rates of expansion are sensitive markers for AD and even more so MCI progression.13 Our group recently developed a computational anatomy technique for fluid registration of parametric mesh models of the lateral ventricles to individual untraced ventricular volumes. To reduce error, ventricles are first traced by hand on a small subset of scans. The results of fluidly mapping these surfaces onto new unlabeled scans are averaged together, greatly reducing segmentation error within each scan. This technique, followed by the radial atrophy mapping approach used earlier to study hippocampal atrophy, rapidly and fully automatically assesses ventricular expansion in neurodegenerative disorders (Fig. 6).61 Studies are currently under way to define ventricular changes in subjects with MCI and AD. A similar approach was used by Carmichael et al.42 who reported that among normal subjects larger ventricles at baseline were predictive of future progression to dementia. These automated surface mapping methods offer the high-throughput required for population-based studies or clinical trials, with N=339 subjects automatically mapped in a recent cohort from the Cardiovascular Health Study.42 These methods provide maps of statistics that are often better able to detect subtle or localized atrophy than traditional numeric summaries. Even so, clinical trials often require neuroimaging outcomes to be expressed in terms of a simple set of numeric summaries (rather than a 3D map, for example). Current efforts are therefore also focused on deriving single numeric summaries from the brain maps that optimally predict conversion from MCI to AD, or therapeutic response in large patient cohorts assessed with longitudinal MRI.79
Dementia with Lewy bodies (DLB) is the second most common dementing disorder affecting the elderly. Patients with DLB frequently show several of the following cardinal features - progressive cognitive decline, early-onset hallucinations and delusions, Parkinsonism and a fluctuating course. The pathologic hallmarks for DLB are synuclein-rich intracellular deposits known as Lewy bodies. Frequently Lewy body and AD-type pathology coincide, and when they do, patients may present with a clinical picture closely resembling AD making it difficult to differentiate the two disorders on clinical grounds alone. In a VBM study, Jack and colleagues48 recently reported a distinct cortical atrophy pattern in DLB that may help differentiate between the two most common dementias — namely hippocampal and inferior temporal preservation along with midbrain atrophy in DLB — when compared to AD subjects. Using the cortical pattern matching technique, Ballmeier et al.80 demonstrated that preservation of the temporal and orbitofrontal cortices in demented subjects is suggestive of DLB as opposed to AD.
Frontotemporal dementia (FTD) is a spectrum of disorders that typically affects younger patients than other dementias. The behavioral variant (also known as frontal variant FTD, fvFTD) is characterized by progressive atrophy of the frontal lobes81 and affects middle-aged adults, while the two temporal variants – primary progressive aphasia and semantic dementia, are characterized by progressive language impairments, a predominant left temporal atrophy pattern82 and a tendency to affect middle- to older-aged adults. Two preliminary studies using surface-based computational anatomy techniques demonstrated a frontal and right-sided predilection of fv FTD versus posterior and left-sided predilection of semantic dementia83 and the classic frontal FTD versus the classic posterior AD-associated cortical atrophy pattern.84
Epilepsy is a chronic neurologic disorder that affects humans throughout their life span. Its main characteristics are episodic spontaneous synchronous neuronal discharges that clinically manifest as seizures. It is another neurological disease with enormous social and emotional impact. The commonest variant is mesial temporal lobe epilepsy (MTLE), caused by aberrant neuronal circuitry most often localized to the hippocampus and/or the amygdala. When therapy-resistant and intractable, MTLE is frequently surgically amenable through a procedure known as anterior temporal lobectomy. Given the need for extreme caution, precise diagnosis and localization of the seizure focus are imperative before any patient is subjected to an invasive neurosurgical procedure. Brain mapping has therefore been increasingly relied upon in clinical decisionmaking. 3D hippocampal maps of seizure-free versus non-seizure-free subjects who underwent temporal lobectomy for MTLE revealed that more severe hippocampal atrophy ipsilateral to the pre-operative seizure origin and contralateral anterior and lateral hippocampal involvement carried a poorer prognosis with respect to a seizure-free post-surgical outcome.85 A surface-based cortical thickness study analyzed the pre-operative MRI scans of seizure-free MTLE patients who underwent anteromesial temporal lobectomy in comparison to age-matched controls. The patient group showed cortical thinning in the frontal poles, frontal opercular, orbitofrontal, lateral temporal and occipital areas. Patients with a longer duration of illness showed greater atrophy of the superior frontal and parahippocampal gyri ipsilaterally.85
The human immunodeficiency virus (HIV) is readily detected in brain tissue of seropositive patients. Sometimes the infection is latent and patients are asymptomatic. Other times they may suffer from dementia, encephalopathy, demyelination, Parkinsonism or other movement disorders, sleep abnormalities and/or opportunistic CNS infections. Using the cortical pattern matching technique, Thompson et al.86 investigated the changes in cortical thickness in 26 nondemented HIV and 14 control patients (Fig. 7). Despite aggressive retroviral therapy, the cortical maps revealed 15% atrophy of the primary sensorimotor and the premotor cortices in HIV subjects relative to the control group. Focal atrophy of the frontopolar and perisylvian regions correlated with the CD4+ T-lymphocyte counts, while prefrontal and parietal cortical thinning correlated with psychomotor slowing and motor deficits. TBM analyses of the same data (Fig. 7) extended these findings by reporting white matter degenerative changes in regions underlying the primary and association sensorimotor cortices. These changes were likewise linked to information processing speed and CD4+ T-cell counts, a measure of immune system integrity.53 A third publication assessed callosal and ventricular changes in a slightly larger HIV-infected group (n=30) against 21 control subjects and reported anterior callosal thinning and frontal ventricular horn enlargement in the seropositive group. Once again, the structural changes (e.g., ventricular expansion) were linked to the CD4+ count and cognitive change.87
Schizophrenia is a chronic psychiatric disorder manifesting with episodic psychotic disturbances and progressive pervasive social, cognitive and functional deterioration. Neuroimaging data have shown the toll of the disease on brain structure to be profound, although there is disagreement regarding its cellular or molecular basis. Schizophrenic patients show hippocampal atrophy88 which maps to the anterior and mid-portions of the CA1 and CA2 subfields.89 This predominantly anterior hippocampal atrophy pattern has been independently replicated by another research group as well.90,91 The disease effect may be lateralized in twins with the left hippocampus being more affected in dizygotic and less affected in monozygotic schizophrenic twins relative to their unaffected siblings.92 Cortical thickness studies have revealed atrophy of the frontal, parietal, temporal and occipital cortices.93,94 Current brain mapping efforts in schizophrenia focus on determining whether these cortical changes are modulated by antipsychotic treatment,95 and to what extent they occur in the prodromal phase of the illness,96 when medications may resist or delay onset of psychosis. The presence of late life depression (illness onset after the age of 60) has been linked to bilateral temporal and parietal atrophy.97 Bipolar adults have demonstrated focal deficits in the orbitofrontal and temporal regions as well as the amygdala in one TBM study.98
In child psychiatry, studies of attention-deficit hyperactivity disorder have shown an anteroposterior dichotomy with less dorsal prefrontal and anterior temporal and more posterior temporal and inferior parietal cortical gray matter.99 Children who subsequently evolve to develop bipolar disorder were reported to have increased left temporal and decreased anterior cingulate cortical thickness.100
Abnormal brain development and stalled or disturbed brain growth can occur in utero and in early childhood from genetic predisposition, infection or toxic influences. Fetal alcohol syndrome is diagnosed in infants who suffered heavy prenatal alcohol exposure. It presents with poor brain and somatic growth, brain and facial malformations and mental retardation. Sowell et al.101 demonstrated that the children with fetal alcohol syndrome had 15% gray matter excess in the perisylvian cortices suggestive of aberrant regional cortical development.101
Williams syndrome is a rare genetic disorder caused by a heterozygous deletion of chromosome 7q11.23. Phenotypically, affected children commonly display dysmorphic facial features, congenital heart and renal malformations and mild mental retardation with a characteristic sparing of visuospatial functions relative to language and memory faculties. The structural brain derangements in Williams syndrome include increased gyrification in temporoparietal regions. Despite the significantly smaller brain, and lower gray and white matter volumes overall, these children had thicker gray matter in the perisylvian regions. This is thought to result from aberrations in cell migration proteins such as elastin during early gyrogenesis.102
Overall, brain mapping techniques are providing substantial new insights into the trajectory of neurodegenerative disease, the factors that resist or promote disease onset, and the linkage between cognitive decline and regionally specific changes in brain structure and function. Subjects with MCI, in particular, are at fivefold increased risk for conversion to AD, and are therefore the focus of many clinical trials to delay or modulate disease progression. Because of this focus, brain mapping has been used to identify quantitative features of hippocampal and cortical anatomy that are altered in MCI, including subtle changes that predict conversion or link with subtle changes in specific cognitive domains.
At the same time, rapid developments in the imaging techniques themselves are driving research into the molecular hallmarks of AD. In this review we have focused on MRI scanning, and methods to analyze structural images of the brain, but new PET radiotracers now show great promise as in vivo markers of NFT and amyloid pathology.103 The computational anatomy techniques described here are likely to be of great value in establishing the trajectory of these biomarkers as the disease spreads throughout the cortex.104
Fusion of different brain mapping modalities with postmortem data is also required to fully relate pathology to structural deterioration, and may help in defining the success of cholinergic or neuroprotective therapies. Ongoing developments in high-field MRI (at field strengths of 7T and higher), and in diffusion tensor imaging, now reveal the integrity and connectivity of the white matter in unprecedented detail. These new imaging approaches bring new sources of power for detecting degenerative illness, and are likely to be invaluable in understanding the impact of neurotherapeutic agents to delay or revert disease progression.
This work was generously supported by NIA K23 AG026803 (jointly sponsored by NIA, AFAR, The John A. Hartford Foundation, the Atlantic Philanthropies, the Starr Foundation and an anonymous donor; to LGA), NIA P50 AG16570 (to LGA and PMT); NIBIB EB01651, NLM LM05639, NCRR RR019771, NIH/NIMH R01 MH071940, NIH/NCRR P41 RR013642 and NIH U54 RR021813 (to PMT).
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.