THE LANCET Neurology
THE LANCET Neurology

The Journal

Info for Authors



The Journal
Current Issue
Contents in full
The leading edge
Reflection and Reaction
Review and opinion
Media reviews
The last word
Back Issues

Search Journal
Log Out

Awards and Announcements
Press Services
General Info
Free Sample
Contact Us

THE LANCET Infectious Diseases
Home The Journal Current Issue Review and opinion
Volume 2, Number 2     01 February 2003

Previous Next

Computer-assisted imaging to assess brain structure in healthy and diseased brains

John Ashburner a , John G Csernansky b , Christos Davatzikos c , Nick C Fox d , Giovanni B Frisoni  Corresponding Author Information Send E-mail to Author e , and Paul M Thompson f

Article outline:

Computational anatomy algorithms
Registration strategies
Cross-sectional methods
Prospective methods
The methods
Whole-brain analysis with voxel-based morphometry
Cortical pattern matching
Diffeomorphic mapping of the hippocampus and other closed brain structures
Brain-boundary shift integral method
Voxel-compression mapping
Clinical findings
Structural differences related to age, sex, and genotype
Morphological signatures of brain diseases
Alzheimer's disease
Frontotemporal versus semantic dementia
Huntington's disease
Parkinson's disease
Other brain disorders
Mapping of disease progression in neurodegenerative diseases
Alzheimer's disease
Computational neuroanatomy for clinical diagnosis
Search strategy and selection criteria

Neuroanatomical structures may be profoundly or subtly affected by the interplay of genetic and environmental factors, age, and disease. Such effects are particularly true in healthy ageing individuals and in those who have neurodegenerative diseases. The ability to use imaging to identify structural brain changes associated with different neurodegenerative disease states would be useful for diagnosis and treatment. However, early in the progression of such diseases, neuroanatomical changes may be too mild, diffuse, or topologically complex to be detected by simple visual inspection or manually traced measurements of regions of interest. Computerised methods are being developed that can capture the extraordinary morphological variability of the human brain. These methods use mathematical models sensitive to subtle changes in the size, position, shape, and tissue characteristics of brain structures affected by neurodegenerative diseases. Neuroanatomical features can be compared within and between groups of individuals, taking into account age, sex, genetic background, and disease state, to assess the structural basis of normality and disease. In this review, we describe the strengths and limitations of algorithms of existing computer-assisted tools at the most advanced stage of development, together with available and foreseeable evidence of their usefulness at the clinical and research level.

The link between brain structure and function has been of interest since the golden age of phrenology in the early 1800s. Advances in neuroscience and neuroimaging have led to an increasing recognition that certain neuroanatomical structures may be affected preferentially by particular diseases. The distribution of structural changes reflects the underlying pathology and may determine the clinical phenomenology. These associations are well exemplified by the differential patterns of atrophy seen in the neurodegenerative disorders.

Neurodegenerative brain diseases mark the brain with morphological signatures; detection of these signs may be useful to improve diagnosis, particularly in diseases for which there are few other diagnostic tools. For example, early and disproportionate hippocampal atrophy in people who have memory complaints points to a diagnosis of Alzheimer's disease. By contrast, focal atrophy of the temporal lobe, frontal lobe, or both, makes Alzheimer's disease less likely and a tauopathy such as Pick's disease more likely. 1 Such preset regional patterns of abnormalities extend to the parkinsonian disorders, in which midbrain atrophy is useful in differentiating progressive supranuclear palsy from idiopathic Parkinson's disease, 2 whereas striatal abnormalities and cerebellar atrophy are more common in multiple-system atrophy. 3

Structural changes provide markers by which to track the biological progression of disease. In 1993, despite a lack of support from clinical variables, interferon β was approved by the US Food and Drug Administration on the basis of MRI data. These data showed a slower increase in white-matter hyperintensities in treated than in untreated patients who had multiple sclerosis. 4

The increasing sophistication of MRI allows neuroanatomical structures to be visualised in vivo with exquisite detail; clinical MRI gives good soft-tissue contrast and high (<1 mm) resolution. This imaging detail can identify cerebral structures—the volume, shape, and tissue chracteristics of which are preferentially affected by a given disease or disorder. Quantitative tools such as volumetric measures based on manually traced regions of interest are extensively used to assess overall size of brain structures in individuals with neurodegenerative disorders. These approaches are unsatisfactory in at least two ways. First, in the early stages of many chronic brain diseases, changes in overall size are generally minimal. For example, despite gross hippocampal atrophy in advanced Alzheimer's disease, in the earliest stages only a small percentage of premorbid volume may be lost. 5–7 Moreover, these pathological effects might be hidden by premorbid brain size, age, sex, or genotype. 8–10 Second, brain structure may differ in multiple brain areas, and without assessment of other areas of the brain, a particular regional loss may lack specificity. Hippocampal atrophy is characteristic of Alzheimer's disease but is also present, although less severely, in other degenerative diseases such as frontotemporal degeneration, 11,12 which also features notable frontal and temporal cortical atrophy. 13

Developments in computer science may offer some help. First, new approaches have become increasingly automated, which avoids error-prone and labour-intensive manual measurements. Second, such algorithms can offer unprecedented precision; they can detect brain-volume differences of 0⋅5% between images from the same individual. 14

The effort to develop such algorithms has been referred to as computational anatomy. 15 The individual algorithms can be classified into two broad groups based on their anticipated use: algorithms devised to detect group differences at one point in time and those devised to detect prospective changes over time. The first group can be especially useful to define disease-specific signatures and to reveal the selective vulnerability of different cell types to different pathologies, which may guide treatment strategies. The second group can be applied to one or more individuals to track disease progression, be it the natural history or as modified by treatment. These tools can also correlate symptoms or functions with involvement of a given structure or neural system. It should be kept in mind that, although an algorithm may have been developed for one purpose, it can also be useful for another. Tools in the two groups have been adapted to analyse individual cases, an issue of great interest to improve clinical diagnosis.

In this review, we describe the strengths and limitations of algorithms of computational anatomy at the most advanced stage of development, together with available and foreseeable evidence of their usefulness at the clinical and research level. We discuss separately tools capable of detecting cross-sectional group differences at a given time and also tools that are able to detect changes in individual people over time ( table ). For the group tools, we also address the methods used for analysis—ie, whole-brain versus region of interest. Computational strategies aimed at assessment of individual cases are in their infancy and will be only briefly discussed.

  Registration Analysis
  Matching Modelled brain variability Manually positioned landmarks Modelled voxel variability Measure of interest
Cross-sectional methods
Ashburner and Friston's method 16 Image to template Global No Intensity Volume
RAVENS 17 Image to template Global and local Yes Intensity Volume
Cortical pattern matching 18 Image to template Local Yes Deformation Volume and shape
Diffeomorphic mapping of hippocampus 19 Template to image Local Yes Deformation Volume and shape

Prospective methods
Brain-boundary shift integral 20 Image to image Global No Intensity Volume
Voxel-compression mapping 21 Image to image Local No Deformation Volume and shape
Table 1. Basic features of registration and analysis in computational neuroanatomy

Computational anatomy algorithms

Most computational anatomy algorithms involve some or all of the following steps: brain extraction, in which brain and non-brain voxels (volume elements represented as pixels on the MR images) are separated; tissue segmentation, in which voxels representing grey and white matter and CSF are separated on the basis of intensity values; spatial normalisation, also called registration, in which the voxels of interest are matched to a template or an earlier scan from the same individual; and statistical comparison of different groups of patients or points in time. Errors can arise in the brain-extraction and segmentation phases; for example, partial volume effects can cause some periventricular white matter to be classified as grey. However, the pivotal step of all methods is registration. Cross-sectional methods match images of interest to a reference stereotactic template (a typical brain or a typical hippocampus, etc) or templates to images, and prospective methods match sequential images of the same patients taken at different times, with use of the first image as the reference ( table 16–21). Most methods permit volumetric measurements, although pure shape variables can also be extracted.

Registration strategies
Registration strategies differ in their scope—ie, analysis of the whole brain or preselected regions of interest—and mathematical approach, which takes into account global or local variability of the brain's size and shape. Many cross-sectional methods that account for global variability are completely automated, whereas those that account for local variability frequently require manually positioned landmarks to precisely match the image to the template. Prospective methods use the complexity of each individual's brain structure to align accurately an individual's serial images.

In the statistical comparisons of groups of patients or time points, the variable of interest can be derived from the grey-scale values of voxels (voxel-based approaches) or from deformation fields (deformation-based approaches). Current voxel-based approaches first remove global or global and local variability, and then modulate voxel intensity according to contraction or expansion during registration. Differences in voxel intensity reflect the amount or concentration of grey or white matter.

Cross-sectional methods
Global variability is removed by the approach developed at the Wellcome Department of Imaging Neuroscience group, London, UK, based on statistical parametric mapping, and referred to here as Ashburner and Friston's method. 16 Global and local variability is removed by a technique developed at the Section of Biomedical Image Analysis at the Department of Radiology, University of Pennsylvania, PA, USA, called regional analysis of volumes examined in normalised space (RAVENS). 17

Local registration removes all the variability in shape of regions of interest between individuals (eg, the cortical mantle or the hippocampus). A deformation field can be computed that stores information on the extent of warping, intensity increase or decrease, or a combination of these, for each voxel to obtain the precise match between the images and the template. Deformation fields can be computed by warping the images of interest to a template or vice versa. Therefore, deformation fields are compared rather than registered images. We will discuss cortical pattern matching, developed at the University of California at Los Angeles, CA, USA 18 and diffeomorphic mapping of the hippocampus developed at Washington University School of Medicine, MO, USA ( table). 15,19

Prospective methods
Image registration can be more straightforward for detection of prospective changes within individuals. The method involves matching an image to others for the same person taken at different times. By contrast with cross-sectional methods, the unique morphology of any individual's brain is not an issue, and the complexity of the brain structure itself is used to achieve accurate matching. Only changes over time are measured, and the complex brain shape and size is assumed to be generally invariant in the same individual. Information on prospective global changes can be obtained by rigidly matching serial scans and subtracting the superimposed images. The difference reflects the volume of brain tissue lost or gained (eg, brain-boundary shift integral 20 or structural image evaluation using normalisation of atrophy). 22,23 Information on local changes can be obtained by use of non-linear registration, which permits compression or expansion of each voxel to obtain a precise registration (voxel-compression method). The resulting deformation fields provide a map (voxel-compression map) of the amount of compression or expansion applied at each voxel. This change reflects the amount of brain tissue or cerebrospinal fluid lost or gained between scans. Typical patterns of change under different conditions (eg, normal ageing vs Alzheimer's disease) can be identified by registration and averaging of individual voxel-compression maps and comparing the resulting averages.

To be done well, all methods need high spatial resolution and clear differentiation between tissue types. Normally, three-dimensional, high-resolution, T1-weighted MRI (spoiled gradient or magnetisation-prepared rapid-acquisition gradient recalled echo) acquired with conventional 1⋅5T MR scanners and 1 mm3 voxels (ideally isotropic) across the cranium provide sufficient detail and contrast.

These methods represent a set of flexible and dynamic rather than strictly defined tools. Algorithms originally developed to detect cross-sectional differences among groups have been successfully used on prospective data and to assess changes in individuals. 24 Most algorithms are undergoing constant refinement and updating by the original developers and users, and details of Ashburner and Friston's are available at in the JISCMAIL archives.

Automatic softwares other than those we discuss in detail have been developed for segmentation and registration. 25,26 These systems automatically assign voxels to several subcortical brain structures (striatum, claustrum, thalamus) and to cortical regions (frontal, temporal, parietal, and occipital) and can be useful for analysis of anatomical differences. Besides cortical pattern matching, additional methods have been developed to measure the thickness of the cortex that can also compare cortical anatomy across individuals and groups. 27

The methods

Whole-brain analysis with voxel-based morphometry
The key distinction between the Ashburner and Friston's approach and RAVENS is a difference in spatial normalisation methods. Ashburner and Friston's method uses linear and non-linear warping algorithms that remove global brain variability; RAVENS uses a very high-dimensional (the number of dimensions is equal to the number of voxels in the template and target scans) three-dimensional warping technique aimed at removal of local and global variability. 28 The two methods modulate voxel intensity in relation to voxel expansion or contraction during registration. 28,29

Registered grey-matter images are smoothed with an 8–12 mm filter. This method yields normally distributed data and allows the use of statistical parametric tools. The statistical approach of Ashburner and Friston's method (statistical parametric mapping) is based on the general linear model, and identifies regions of tissue with increased or decreased density or concentration that are significantly related to the effects under study. 30 Ideally, the threshold for significance should be set at p<0⋅05, corrected for multiple comparisons, but when a hypothesis has been formed of the expected effect, a threshold of p<0⋅001 uncorrected can be used. However, like every statistical test, the larger the effect size and group size, the higher the sensitivity of the method for identifying differences. If non-parametric modelling is desirable, a non-parametric version (statistical non-parametric mapping) can be used. 31

Ashburner and Friston's method has been implemented in software that runs under Microsoft Windows or UNIX systems, which can be downloaded from at no cost. The RAVENS method can be used in an automated system, requiring UNIX or Linux operating systems, or a semi-automated mode, and requires UNIX or Linux with OpenGL graphics capabilities, and is available from one of the authors (CD).

Cortical pattern matching
Cortical pattern matching is a sensitive approach that measures the topological variability of the cortex. 32–35 The approach consists of two basic steps: cortical flattening and sulcal matching. To make an average cortical model for a group of individuals, all MR scans are first globally aligned to a standard three-dimensional coordinate space. From each individual's MR scan, a three-dimensional cortical surface model is extracted, consisting of a network of discrete triangular tiles. All consistent sulcal or gyral landmarks are identified on the cortical model, which is then flattened. Sulcal features are aligned across individuals with a warping technique. An average set of sulcal curves derived from many people is overlaid on the flat map. These maps are warped so that those individual sulcal features in them are driven into correspondence with the average set of sulcal curves. These warped images are averaged together and decoded. Importantly, measures of grey-matter density, and functional imaging data if available, may be convected along with these warps and plotted on the average cortex before statistical analysis. As in Ashburner and Friston's method, ANCOVA can be applied to the cortical surface maps. Regions are identified in which deficits are most closely correlated with symptoms or disease progression.

This modelling technique can be executed on high-end desktop computers. These algorithms are frequently used in client–server mode, connecting to a supercomputer for very large-scale analyses, 36 and are available from one of the authors (PT).

Diffeomorphic mapping of the hippocampus and other closed brain structures
Diffeomorphic brain mapping represents neuroanatomical structures by use of templates and their variability by probabilistic transformations applied to the templates. 15,37,38 Displacement maps are generated by calculation of the difference between the locations of several thousand points on the hippocampal surface compared with the overall mean of the points (the template). The transformations are high-dimensional as well as diffeomorphic (invertible, continuous, and differentiable), which makes them especially sensitive to subtle forms of neuroanatomical variation.

When the template is constructed from an MR scan and information identifying the boundary of a structure is carried along with the transformation, selected structures within MR scans of individuals are automatically defined. More importantly, the transformation fields and metrics derived from them by singular value decomposition (ie, non-zero vectors, or eigenvectors) can be used to: define composite shapes and volumes of brain structures in groups of patients, including patterns of hemispherical asymmetry; assess the degree to which the volume and shape of such structures change within individuals over time; and test for differences in brain-structure volumes and shapes between groups with and without neurological diseases. 19,39–41

Definition of the brain structures of interest within the template scan depends on contributions from specialists in neuroanatomy, radiology, and neuropathology. In addition, the initial step of the algorithm is guided by landmarks manually placed within the template and target MR scans. 42 However, once the chosen structures within the template have been defined and the target scans have been landmarked, the burden for applying this neuroanatomical information to large numbers of individual target scans lies solely with the transformations.

Algorithms can be done with software for desktop personal computers. Interested investigators can inquire at Surgical Navigation Technologies (Louisville, Colorado, USA) about obtaining available versions of these methods.

Brain-boundary shift integral method
The comparison of serial images always starts with registration. To assess global cerebral loss, this matching is done by rigid-body registration. As a result of the structural readjustments that occur within the brain in neurodegenerative disorders, the image produced by subtracting the registered repeat image from the initial image shows intensity loss at tissue boundaries. The intensity loss corresponds to the positional shift caused by the structural readjustment, and, hence, tissue loss can be approximated by calculation of the volume through which the boundary has shifted (brain-boundary shift integral). 20

Voxel-compression mapping
Rigid-body registration, however, cannot model localised brain changes. To assess these changes, matching needs to be done with a non-linear algorithm that will permit localised deformation, such as one that models the brain as a compressible viscous fluid. 21,43 The repeat image is first rigidly registered to the initial scan, then non-linear registration is used to match the scans locally as well as globally: the governing fluid equations are repeatedly solved to keep to a minimum all differences between scans. The expansion or contraction that has occurred in each voxel is calculated from the deformation field to create the voxel-compression map; contraction implies local atrophy and expansion implies local growth (eg, in a CSF space).

To find out which features are consistent for a particular disease, robust group analysis is required. Statistical parametric mapping analysis 30 of fluidly registered images incorporates the neuroanatomical sensitivity of within-individual registration and the power of group comparisons done with between-individual registration.

A practical difficulty with any prospective method is that MR scanners regularly undergo substantial hardware and software modifications. Since these changes can substantially alter image quality and intensity contrast, the same scanner and MR acquisition parameters should be used for baseline and repeat scans. As with any other method of measuring changes in brain volume over time, these methods are also sensitive to subtle brain changes that are not disease-related, such as hydration. 44

Other prospective methods have been developed such as structural image evaluation using normalisation of atrophy, 45 which finds all brain-surface edge points and estimates the motion of these edge points from one time point to the next. A cross-sectional evolution has been developed. 22,23 There are no studies that compare the performance of this method and the brain-boundary shift integral method.

Clinical findings

Structural differences related to age, sex, and genotype
By application of Ashburner and Friston's method to a large database of normal individuals aged 18–79 years, age-related grey-matter loss could be seen bilaterally in the insula, superior parietal gyri, and central and cingulate sulci, whereas the amygdala, hippocampi, and entorhinal cortex exhibited little or no age effect. 29 On images taken in 91 people aged 59–85 years (mean age 80⋅7) enrolled in the Baltimore Longitudinal Study of Aging, 46 RAVENS detected age-related grey-matter loss in the hippocampus and in the orbitofrontal cortex ( figure 1 ). Some of these findings are in agreement with pathological data, but future studies will need to clarify whether inconsistent results are due to different performances of the image analysis tools or to different study populations. In this dataset, RAVENS also showed increased white-matter volume in the anterior portion of the splenium in women but not in men. Significant correlations were noted between regional callosal measurements and verbal and non-verbal neuropsychological tests. 47 These results are consistent with post-mortem reports 48 and with the hypothesis that greater splenial size facilitates more bilateral processing of tasks by women.

Figure 1. Age-related changes of brain structure. Top: percentage difference between younger (59–69 years) and older (75–90 years) participants, revealing regional patterns of brain atrophy in the grey matter (left) and the corpus callosum (right). The underlying fuzzy grey-scale images are sections of the average distribution of the grey matter from all participants. Bottom: projection of the age differences on the cortical surface. Colour-scale shows relative change as a decimal fraction—eg, 0⋅02=2%.

In fraternal (dizygotic) and identical (monozygotic) twins, cortical pattern matching has been used to investigate the amount and location of genetic control over the cerebral cortex ( figure 2 ). The patterns were averaged and compared with the average differences that would be seen between pairs of randomly selected unrelated individuals. Fraternal twins exhibit 30% of the normal between-individual differences, and these affinities are largely restricted to perisylvian language and spatial association cortices. Genetically identical twins display only 10–30% of normal differences in a large anatomical band spanning frontal, sensorimotor, and Wernicke's language cortices, which suggests strong genetic control of brain structure in these regions. 33

Figure 2. Genetic continuum of similarity in brain structure in identical and fraternal twins, and non-related individuals. Colour-coded maps show the percentage reduction in within-pair variance for each cortical region. Purple/blue=normal between-individual difference (100% of normal); green=60% of normal differences; red=30% of normal differences. F=frontal cortex; S/M=sensorimotor cortex; W=Wernicke's language cortex. Figure reproduced with permission from Nature Publishing Group . 33

Morphological signatures of brain diseases

Alzheimer's disease
With Ashburner and Friston's method, hippocampal atrophy has been visualised in Alzheimer's disease. This method and cortical pattern matching have shown volume loss in the posterior cingulate gyrus and adjacent precuneus, and the temporoparietal association cortex. 49,50 This atrophy profile is consistent with metabolic deficits commonly reported in fluorodeoxyglucose PET studies, and with amyloid-β peptide and neurofibrillary-tangle distribution. 51,52 In most studies in Alzheimer's disease in which Ashburner and Friston's method was used, 53–55 atrophy has been noted in the caudate nucleus. Although this finding is not immediately understandable, it highlights the need for strong biological or physiological plausibility to accept the findings of computational neuroanatomy techniques. In three patients who had very mild Alzheimer's disease (Mini-Mental State Examination scores of 26 and 27), compared with 26 non-demented controls, the amygdalar-hippocampal complex was significantly atrophied bilaterally (p<0⋅0001 uncorrected for multiple comparisons), which suggests that the technique is also sensitive in small groups and in the earliest stages of the disease ( figure 3 ). 54

Figure 3. Early grey-matter changes in Alzheimer’s disease. Voxels of decreased grey-matter density detected with Ashburner and Friston’s method in the amygdala and hippocampal complex and temporal cortex are shown in yellow. Significant voxels are superimposed on MRI of one participant. Figure reproduced with permission from BMJ Publishing Group. 54

With use of diffeomorphic brain mapping of the hippocampus, volume loss has been detected in the head of the hippocampus and along the lateral edge of the hippocampal body in patients with very mild Alzheimer's disease, but in non-demented elderly controls a general flattening of the structure without volume loss was detected ( figure 4 ). The linear combination of hippocampal volumes with two eigenvectors, which reflects a distinct pattern of neuroanatomical deformation, significantly distinguished individuals with very mild Alzheimer's disease from non-demented elderly controls, with sensitivity and specificity of 83% and 78%. A different set of eigenvectors, which reflect a different pattern, distinguished the non-demented elderly from younger controls with 100% accuracy. Therefore, the known biological and pathological differences between ageing and Alzheimer's disease can be appreciated at the macroscopic structural level as different shape deformations of selected brain structures.

Figure 4. Shape changes of the hippocampus in Alzheimer’s disease (AD). The displacement map shows deformation of the right hippocampus (volume loss in head of hippocampus and along lateral extent of hippocampal body) in patients with very mild AD compared with healthy older people (left) and in healthy older compared to younger people (structure flattening with no volume loss, right). The magnitude and direction of displacement are coded by a flame scale: blue to purple=inward deformities consistent with local volume loss.

Frontotemporal versus semantic dementia
Different patterns of cortical atrophy have been seen in frontotemporal and semantic dementia by use of Ashburner and Friston's method. In frontotemporal dementia, there is notable involvement of the right dorsolateral frontal and left premotor cortex, and in semantic dementia involvement of the anterior temporal cortex and the amygdala and anterior hippocampal region can be seen bilaterally; the ventromedial frontal and left anterior cingulate cortex, the posterior orbital frontal regions, and the insula, bilaterally, are involved in both these types of dementia. 56 These results suggest broad involvement of the limbic system in these two disorders, but with greater frontal lobe involvement in frontotemporal and more temporal lobe loss in semantic dementia. Whether the biological basis of these findings is differential involvement of neural networks is unclear, but computational neuroanatomy methods make it possible to address this question.

Huntington's disease
Ashburner and Friston's technique has revealed grey-matter pathology in the striatum and other cortical and subcortical structures in asymptomatic individuals who test positive for genetic risk of Huntington's disease. Therefore, the presence and progression of structural pathology can probably be detected and monitored in asymptomatic patients with Huntington's disease, with potential prognostic and therapeutic implications. Moreover, the pattern of cortical involvement in patients with Huntington's disease affects mainly the posterior regions, which challenges the notion that cognitive disturbances are due to involvement of frontostriatal circuits. 57

Parkinson's disease
Atrophic changes and volumetric increases of some structures can be detected in patients with neurodegenerative disorders. In idiopathic Parkinson's disease, locally increased grey-matter concentration has been seen in the nucleus ventralis intermedius of the thalamus contralateral to the tremor side. Raised grey-matter concentration was significantly correlated with increased tremor amplitude. 58 Whether this correlation is a primary phenomenon or secondary to basal ganglia pathology is unclear, but the finding suggests that the structural brain changes of neurodegenerative disorders are more complex than mere progressive loss.

Other brain disorders
Temporal-lobe epilepsy, schizophrenia, and drug addiction have also been studied with Ashburner and Friston's method, 59–61 and other disorders are likely to follow suit. In patients who have temporal-lobe epilepsy, prefrontal grey-matter atrophy was noted in addition to the well-known hippocampal atrophy, and might be the biological cause for the executive dysfunction frequently described in epileptic patients. 59

Strikingly accelerated cortical grey-matter loss has been seen with cortical pattern matching in young schizophrenic patients well after the onset of psychosis, possibly reflecting the convergence of genetic and non-genetic factors on the pathogenesis of the disease. 62 By use of diffeomorphic mapping of the hippocampus, a shape deformity has been identified in schizophrenia distinct from that associated with either dementia or ageing and localised to the head of the hippocampus. 19,63

Decreased grey-matter concentration has been seen in the orbitofrontal, cingulate, and temporal cortices in cocaine users. 61 Structural damage is positively correlated with the duration of drug use, which suggests direct neurotoxic effects. 64

Mapping of disease progression in neurodegenerative diseases

Alzheimer's disease
Rates of global brain atrophy based on the brain-boundary shift integral are substantially higher in 18 patients with Alzheimer's disease than in 31 age-matched controls with a scan interval as short as 1 year. 14 Good separation was seen between the groups ( figure 5 ); patients with Alzheimer's disease had annual rates of atrophy of 2⋅8% (SD 0⋅9) and controls had rates of 0⋅2% (0⋅3).

Figure 5. Rates of global brain-volume loss in Alzheimer’s disease from serial MRI. AD=Alzheimer’s disease. Rates are computed with the brain-boundary shift integral method. Reproduced with permission from John Wiley and Sons Inc. 14

Power calculations have been done to estimate sample sizes required to monitor treatment effects in clinical trials of drugs aimed to slow the progression of neurodegenerative diseases. These calculations have shown that atrophy rates derived from the brain-boundary shift integral would require smaller numbers (18 patients per treatment group with a magnitude of treatment effect of 50%) than would other techniques relying on segmentation of regional structures (42–187 patients). 65 However, since rates of total brain atrophy are not specific for a particular degenerative disease (eg, Alzheimer's disease and frontotemporal dementia have similar global atrophy rates, figure 6 ), 66 these rates have limited application in the differential diagnosis of dementia. With voxel-compression methods, characteristic patterns have been shown in the different dementias, with Alzheimer's disease featuring diffuse atrophy, in contrast to the focal anterior atrophy of frontotemporal dementia. These patterns are consistent with the clinical symptoms of the disease, as well as post-mortem evidence. 67 This technique is sensitive to early change and can identify regional brain atrophy before clinical diagnosis in Alzheimer's disease and frontotemporal dementia. 68,69

Figure 6. Rates of global and local atrophy in Alzheimer’s disease and frontotemporal dementia from serial MRI. Coronal MRI with voxel-compression-mapping colour overlays. Top: change over 11 months in healthy control. Middle: change over 14 months in an individual with Alzheimer’s disease. Bottom: change over 15 months in an individual with frontotemporal dementia. Reproduced with permission from Elsevier Science. 66

Although not originally devised for prospective analyses, Ashburner and Friston's method and cortical pattern matching can also be used to analyse serially acquired images. In each case, registration is done on the baseline images rather than the template, and the registered time series from individual patients are subsequently aligned to a common template. In a study of 17 patients with Alzheimer's disease assessed by prospective methods, 36 cortical pattern matching showed progression of atrophy from the medial temporal to the temporoparietal and frontal grey matter, whereas sensorimotor regions were generally spared. The region of deficits spread centrifugally from cingulate and other limbic regions into frontal cortices, with greatest neocortical atrophy detected in the left hemisphere. This progression occurred over a 2⋅5-year period in which Mini-Mental State Examination scores declined from 18 to 13. The extent of local grey-matter loss correlated with patients’ declining cognitive scores. Moreover, different brain structures had different rates of change. Although the cortex loses grey matter at a rate of 4–5% per year locally, the inferior ventricular horns have greatest dynamic change rates (15–16% per year). In these regions, expansion rates are bilaterally significant even in controls (2–4% per year). 36

Computational neuroanatomy for clinical diagnosis
The clinical findings discussed so far have shown structural differences or correlations in groups of individuals. To be applicable to clinical practice, the techniques need to be sensitive to quantitative, qualitative, or both, departures from normal brain structure in an individual.

The brain-boundary shift integral technique has been used to detect atrophy progression in a cognitively intact patient with preclinical Alzheimer's disease. Atrophy progression was detected 5 years before the patient fulfilled clinical criteria for probable disease and more than 2 years before the onset of any symptom. 69 In the same patient, voxel-compression mapping located the spread of atrophy from within the medial temporal to temporoparietal regions. 69

The main drawback of the brain-boundary shift integral and voxel-compression methods is that diagnosis cannot be made on first observation of the patient, but only after a follow-up scan has been done. This time lapse is undesirable if the patient already has symptoms at the first observation, and a diagnosis is urgently needed to start treatment. Therefore, methods sensitive to brain defects on one examination are needed.

Although not originally devised for analysis in individuals, Ashburner and Friston's method has been applied with minor modifications of the methods (smoothing at 12–14 mm, significance threshold at p<0⋅05 uncorrected for multiple comparisons) with some success. In all ten patients with gross malformations of cortical development, the method was sensitive to changes of cortical grey-matter content and occasionally better than visual inspection ( figure 7 ). 70

Figure 7. Neocortical grey matter in a patient with right-sided polymicrogyria. Top: the area of substantial grey-matter increase. Bottom: decrease computed with Ashburner and Friston’s method adapted for individual case analysis. The changes in the right hemisphere (on the right side of the image) could be appreciated on visual inspection, but contralateral changes of decreased grey matter were identified only with computerised analysis. Figure reproduced with permission from Elsevier Science. 70

Although these results support the potential of Ashburner and Friston's approach in single-case analysis, the method may not yet be optimum for actually making a diagnosis and needs to be studied in much greater depth before it can be transferred to clinical practice.

Diffeomorphic mapping of the hippocampus has been used to separate patients with mild Alzheimer's disease from non-demented elderly controls based on shape functions using discriminant analysis. 39


Computational neuroanatomy is a rapidly expanding area of research driven by advances in computing power, imaging technology, and algorithm development. In the past 10 years, several methods have been developed to investigate cerebral pathology in living patients and to probe brain development and ageing. These methods capture the extraordinary morphological variability of the human brain through mathematical models sensitive to subtle changes in neuroanatomical shape, complexity, and tissue characteristics. All of these changes may be produced by neurodegenerative diseases; these features can be compared within and between groups of individuals who differ for age, sex, disease state, or genetic background. Multidimensional maps of normal and diseased states are being built that integrate these structural signatures of disease with functional, neurophysiological, and genetic data. These techniques are already being used as surrogate outcome measures to detect decreased tissue loss in trials of drugs designed to slow the progression of Alzheimer's disease. 14

Available techniques might also help clinical diagnosis and management of patients. Visual inspection of, for example, voxel-compression maps provides a diagnostic tool to assist in, although not provide, clinical diagnosis. Similarly, uncorrected t statistic images from statistical parametric mapping may also provide similar qualitative information. Rates of brain-tissue loss might help to distinguish normal ageing from patients with Alzheimer's disease and lead to improved assessment of drug efficacy in the clinic. However, the aim of a fully computerised diagnosis of neurodegenerative diseases—albeit theoretically feasible—is still to be achieved. Techniques that give a measure of similarity between an individual patient and disease maps will need to be developed. The integration of structural with functional imaging 71,72 and MR spectroscopy 73,74 into multidimensional maps might further improve diagnostic accuracy. The momentum of this area of research is such that significant developments are expected in the near future.

Search strategy and selection criteria

The algorithms discussed included in this review were chosen, based on the authors’ knowledge, by GBF and on a PubMed search with the following query syntax: “(morphometry[ti] OR structure[ti] OR mapping[ti]) AND (brain OR cerebral) AND (magnetic resonance OR MR OR MRI)”.

Authors’ contributions

All authors except GBF wrote the sections of the review pertinent to the method developed in their laboratory. GBF drafted the structure of the original and revised versions of the review and edited the other authors’ contributions. All authors revised and contributed to the final version.

Conflict of interest

JA, CD, GBF, and PMT have no conflicts of interest. JGC, in conjunction with Washington University and Medtronics Inc, holds a patent on a method for the neuromorphometry-based diagnosis of neuropsychiatric disorders. NCF has received financial support for imaging research from GlaxoSmithKline, Novartis, Janssen, and Elan.

Role of the funding source

JA is funded by the Wellcome Trust. JGC's contribution to this article was supported by Public Health Service (USA) grants MH/AG 60883 and MH 62130 and CD's was supported in part by NIH grant R01-AG-14971 and NIH contract AG-93-07. NCF is supported by the MRC (UK) through a Senior Clinical Fellowship, GBF by grants from the Italian national health service RF 00.343 from the Archivio Normativo Italiano di Morfometria Cerebrale con Risonanza Magnetica, and RA 00.61 from Decadimento Cognitivo Lieve non Dementigeno: Stadio Preclinico di Malattia di Alzheimer e Demenza Vascolare, and PMT by grants from the National Center for Research Resources (P41 RR13642), the National Library of Medicine (LM/MH05639), the National Institute of Neurological Disorders and Stroke and the National Institute of Mental Health (NS38753), and by the Human Brain Project (P20 MH/DA52176). No funding source had any influence over the preparation of the manuscript or any part in the decision to submit the paper for publication.


JGC thanks Michael Miller, Ulf Grenander, Sarang Joshi, and Lei Wang for development of the methods for diffeomorphic brain mapping. CD thanks Susan Resnick for her contribution to the validation studies of the RAVENS method. NCF thanks Rachael Scahill and Emma Lewis for help in drafting of the paper, GBF Cristina Testa and Francesca Sabattoli for help in the editing of the paper, and PMT, Arthur Toga, Michael Mega, Kiralee Hayashi, Andrew Janke, and Greig de Zubicaray for help in the research at UCLA.

1 Scheltens P, Fox N, Barkhof F, De Carli C. Structural magnetic resonance imaging in the practical assessment of dementia: beyond exclusion. Lancet Neurol 2002; 1: 13-21. [PubMed]

2 Warmuth-Metz M, Naumann M, Csoti I, Solymosi L. Measurement of the midbrain diameter on routine magnetic resonance imaging: a simple and accurate method of differentiating between Parkinson disease and progressive supranuclear palsy. Arch Neurol 2001; 58: 1076-79. [PubMed]

3 Bhattacharya K, Saadia D, Eisenkraft B, et al. Brain magnetic resonance imaging in multiple-system atrophy and Parkinson disease: a diagnostic algorithm. Arch Neurol 2002; 59: 835-42. [PubMed]

4 Paty DW, Li DK. Interferon beta-1b is effective in relapsing-remitting multiple sclerosis, II: MRI analysis results of a multicenter, randomized, double-blind, placebo-controlled trial. UBC MS/MRI Study Group and the IFNB Multiple Sclerosis Study Group. Neurology 1993; 43: 662-67. [PubMed]

5 Du AT, Schuff N, Amend D, et al. Magnetic resonance imaging of the entorhinal cortex and hippocampus in mild cognitive impairment and Alzheimer's disease. J Neurol Neurosurg Psychiatry 2001; 71: 441-47. [PubMed]

6 Jack CR, Petersen RC, Xu YC, et al. Prediction of AD with MRI-based hippocampal volume in mild cognitive impairment. Neurology 1999; 52: 1397-403. [PubMed]

7 Convit A, de Asis J, de Leon MJ, Tarshish CY, De Santi S, Rusinek H. Atrophy of the medial occipitotemporal, inferior, and middle temporal gyri in non-demented elderly predict decline to Alzheimer's disease. Neurobiol Aging 2000; 21: 19-26. [PubMed]

8 Coffey CE, Wilkinson WE, Parashos IA, et al. Quantitative cerebral anatomy of the aging human brain: a cross-sectional study using magnetic resonance imaging. Neurology 1992; 42: 527-36. [PubMed]

9 Cohen RM, Small C, Lalonde F, Friz J, Sunderland T. Effect of apolipoprotein E genotype on hippocampal volume loss in aging healthy women. Neurology 2001; 57: 2223-28. [PubMed]

10 Tohgi H, Takahashi S, Kato E, et al. Reduced size of right hippocampus in 39- to 80-year-old normal subjects carrying the apolipoprotein E epsilon4 allele. Neurosci Lett 1997; 236: 21-24. [PubMed]

11 Jack CR, Petersen RC, O'Brien PC, Tangalos EG. MR-based hippocampal volumetry in the diagnosis of Alzheimer's disease. Neurology 1992; 42: 183-88. [PubMed]

12 Frisoni GB, Laakso MP, Beltramello A, et al. Hippocampal and entorhinal cortex atrophy in frontotemporal dementia and Alzheimer's disease. Neurology 1999; 52: 91-100. [PubMed]

13 Boccardi M, Laakso MP, Bresciani L, et al. The MRI pattern of frontal and temporal brain atrophy in frontotemporal dementia. Neurobiol Aging 2003; 24: 95-103. [PubMed]

14 Fox NC, Freeborough PA. Brain atrophy progression measured from registered serial MRI: validation and application to Alzheimer's disease. J Magn Reson Imaging 1997; 7: 1069-75. [PubMed]

15 Grenander U, Miller MI. Computational anatomy: an emerging discipline. Q Appl Math 1998; 4: 617-94. [PubMed]

16 Ashburner J, Friston KJ. Voxel-based morphometry: the methods. Neuroimage 2000; 14: 805-21. [PubMed]

17 Davatzikos C, Genc A, Xu D, Resnick SM. Voxel-based morphometry using the RAVENS maps: methods and validation using simulated longitudinal atrophy. Neuroimage 2001; 14: 1361-69. [PubMed]

18 Thompson PM, Mega MS, Toga AW. Disease-specific brain atlases. In: Toga A.W. , Mazziotta J.C. (Eds.) 2000; Brain mapping: the disorders pp. 131-77. London: Academic Press

19 Csernansky JG, Joshi S, Wang L, et al. Hippocampal morphometry in schizophrenia via high dimensional brain mapping. Proc Natl Acad Sci USA 1998; 95: 11406-11. [PubMed]

20 Freeborough PA, Fox NC, Kitney RI. Interactive algorithms for the segmentation and quantitation of 3-D MRI brain scans. Comput Methods Programs Biomed 1997; 53: 15-25. [PubMed]

21 Freeborough PA, Fox NC. Modeling brain deformations in Alzheimer's disease by fluid registration of serial 3D MR images. J Comput Assist Tomogr 1998; 22: 838-43. [PubMed]

22 Smith SM, Zhang Y, Jenkinson M, et al. Brain atrophy analysis using single- and multiple-time-point data: FMRIB technical report TR01SMS1. (accessed Nov 18, 2002).

23 Stevenson VL, Smith SM, Matthews PM, Miller DH, Thompson AJ. Monitoring disease activity and progression in primary progressive multiple sclerosis using MRI: sub-voxel registration to identify lesion changes and to detect cerebral atrophy. J Neurol 2002; 249: 171-77. [PubMed]

24 Ashburner J, Friston KJ. Why voxel-based morphometry should be used. Neuroimage 2001; 14: 1238-43. [PubMed]

25 Collins D, Holmes C, Peters T, Evans A. Automatic 3D model-based neuroanatomical segmentation. Hum Brain Mapp 1995; 3: 190-208. [PubMed]

26 Zijdenbos AP, Evans AC, Riahi F, Sled J, Chui J, Kollokian V. Automatic quantification of multiple sclerosis lesion volume using stereotaxic space. Proceedings of the 4th International Conference on Visualization in Biomedical Computing, Hamburg, September 1996: 439–48.

27 Fischl B, Dale AM. Measuring the thickness of the human cerebral cortex from magnetic resonance images. Proc Natl Acad Sci USA 2000; 97: 11050-55. [PubMed]

28 Shen D, Davatzikos C. HAMMER: hierarchical attribute matching mechanism for elastic registration. IEEE Trans Med Imaging (in press).

29 Good CD, Johnsrude IS, Ashburner J, Henson RN, Friston KJ, Frackowiak RS. A voxel-based morphometric study of ageing in 465 normal adult human brains. Neuroimage 2001; 14: 21-36. [PubMed]

30 Friston KJ, Holmes AP, Worsley KJ, Poline JB, Frith CD, Frackowiak RSJ. Statistical parametric maps in functional imaging: a general linear approach. Hum Brain Mapp 1995; 2: 189-210. [PubMed]

31 Nichols TE, Holmes AP. Nonparametric permutation tests for functional neuroimaging: a primer with examples. Hum Brain Mapp 2002; 15: 1-25. [PubMed]

32 Thompson PM, Toga AW. A surface-based technique for warping 3-dimensional images of the brain. IEEE Trans Med Imaging 1996; 15: 1-16. [PubMed]

33 Thompson PM, Cannon TD, Narr KL, et al. Genetic influences on brain structure. Nat Neurosci 2001; 4: 1253-58. [PubMed]

34 Davatzikos C. Spatial normalization of 3D brain images using deformable models. J Comput Assist Tomogr 1996; 20: 656-65. [PubMed]

35 Fischl B, Sereno MI, Tootell RBH, Dale AM. High-resolution inter-subject averaging and a coordinate system for the cortical surface. Hum Brain Mapp 1999; 8: 272-84. [PubMed]

36 Thompson PM, Hayashi KM, de Zubicaray G, et al. Detecting dynamic and genetic effects on brain structure using high-dimensional cortical pattern matching. Proceedings of the International Symposium on Biomedical Imaging, Washington, DC, 2002 July 7–10.

37 Grenander U. General pattern theory. 1993; London: Oxford University Press

38 Miller MI, Christensen GE, Amit Y, Grenander U. Mathematical textbook of deformable neuroanatomies. Proc Natl Acad Sci USA 1993; 90: 11944-48. [PubMed]

39 Csernansky JG, Wang L, Joshi S, et al. Early DAT is distinguished from aging by high dimensional mapping of the hippocampus. Neurology 2000; 55: 1636-43. [PubMed]

40 Miller MI, Banerjee A, Christensen G, et al. Statistical methods in computational anatomy. Stat Methods Med Res 1997; 6: 267-99. [PubMed]

41 Wang L, Joshi SC, Miller MI, Grenander U, Csernansky JG. Statistical analysis of hippocampal asymmetry. Neuroimage 2001; 14: 531-45. [PubMed]

42 Haller JW, Banerjee A, Christensen GE, et al. Three-dimensional hippocampal morphometry by high dimensional transformation of a neuroanatomical atlas. Radiology 1997; 202: 504-10. [PubMed]

43 Christensen GE, Rabbitt RD, Miller MI. Deformable templates using large deformation kinematics. IEEE Trans Image Proc 1996; 5: 1435-47. [PubMed]

44 Walters RJ, Fox NC, Crum WR, Taube D, Thomas DJ. Haemodialysis and cerebral oedema. Nephron 2001; 87: 143-47. [PubMed]

45 Smith SM, De Stefano N, Jenkinson M, Matthews PM. Normalised accurate measurement of longitudinal brain change. J Comput Assist Tomogr 2001; 25: 466-75. [PubMed]

46 Resnick SM, Goldszal AF, Davatzikos C, et al. One-year age changes in MRI brain volumes in older adults. Cereb Cortex 2000; 10: 464-72. [PubMed]

47 Davatzikos C, Resnick SM. Sex differences in anatomic measures of interhemispheric connectivity: correlations with cognition in women but not in men. Cereb Cortex 1998; 8: 635-40. [PubMed]

48 DeLacoste-Utamsing C, Holloway RL. Sexual dimorphism in the human corpus callosum. Science 1982; 216: 1431-32. [PubMed]

49 Good CD, Scahill RI, Fox NC, et al. Automatic differentiation of anatomical patterns in the human brain: validation with studies of degenerative dementias. Neuroimage 2002; 17: 29-46. [PubMed]

50 Thompson PM, Mega MS, Woods RP, et al. Cortical change in Alzheimer's disease detected with a disease-specific population-based brain atlas. Cereb Cortex 2001; 11: 1-16. [PubMed]

51 Mega MS, Chen S, Thompson PM, et al. Mapping pathology to metabolism: coregistration of stained whole brain sections to PET in Alzheimer's disease. Neuroimage 1997; 5: 147-53. [PubMed]

52 Mega MS, Chu T, Mazziotta JC, et al. Mapping biochemistry to metabolism: FDG-PET and amyloid burden in Alzheimer's disease. Neuroreport 1999; 10: 2911-17. [PubMed]

53 Baron JC, Chetelat G, Desgranges B, et al. In vivo mapping of gray matter loss with voxel-based morphometry in mild Alzheimer's disease. Neuroimage 2001; 14: 298-309. [PubMed]

54 Frisoni GB, Testa C, Zorzan A, et al. Detection of grey matter loss in mild Alzheimer's disease with voxel based morphometry. J Neurol Neurosurg Psychiatry 2002; 73: 657-64. [PubMed]

55 Rombouts SA, Barkhof F, Witter MP, Scheltens P. Unbiased whole-brain analysis of gray matter loss in Alzheimer's disease. Neurosci Lett 2000; 285: 231-33. [PubMed]

56 Rosen HJ, Gorno-Tempini ML, Goldman WP, et al. Patterns of brain atrophy in frontotemporal dementia and semantic dementia. Neurology 2002; 58: 198-08. [PubMed]

57 Rosas HD, Liu AK, Hersch S, et al. Regional and progressive thinning of the cortical ribbon in Huntington's disease. Neurology 2002; 58: 695-701. [PubMed]

58 Kassubek J, Juengling FD, Hellwig B, Spreer J, Lucking CH. Thalamic gray matter changes in unilateral Parkinsonian resting tremor: a voxel-based morphometric analysis of 3-dimensional magnetic resonance imaging. Neurosci Lett 2002; 323: 29-32. [PubMed]

59 Keller SS, Mackay CE, Barrick TR, Wieshmann UC, Howard MA, Roberts N. Voxel-based morphometric comparison of hippocampal and extrahippocampal abnormalities in patients with left and right hippocampal atrophy. Neuroimage 2002; 16: 23-31. [PubMed]

60 Hulshoff Pol HE, Schnack HG, Mandl RC, et al. Focal gray matter density changes in schizophrenia. Arch Gen Psychiatry 2001; 58: 1118-125. [PubMed]

61 Franklin TR, Acton PD, Maldjian JA, et al. Decreased gray matter concentration in the insular, orbitofrontal, cingulate, and temporal cortices of cocaine patients. Biol Psychiatry 2002; 51: 134-42. [PubMed]

62 Thompson PM, Vidal CN, Giedd JN, et al. Mapping adolescent brain change reveals dynamic wave of accelerated gray matter loss in very early-onset schizophrenia. Proc Natl Acad Sci USA 2001; 98: 11650-55. [PubMed]

63 Csernansky JG, Wang L, Jones D, et al. Hippocampal deformities in schizophrenia characterized by high dimensional brain mapping. Am J Psychiatry 2002; 159: 1-7. [PubMed]

64 O'Neill J, Cardenas VA, Meyerhoff DJ. Separate and interactive effects of cocaine and alcohol dependence on brain structures and metabolites: quantitative MRI and proton MR spectroscopic imaging. Addict Biol 2001; 6: 347-61. [PubMed]

65 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. [PubMed]

66 Janssen JC, Scahill RI, Schott JM, et al. In vivo mapping of neurodegeneration in frontotemporal lobar degeneration. Proceedings of the 8th International Conference of Alzheimer's Disease and Related Disorders, Stockholm, Sweden, 2002 July 20–25: S479.

67 Chan D, Fox NC, Jenkins R, Scahill RI, Crum WR, Rossor MN. Rates of global and regional cerebral atrophy in AD and frontotemporal dementia. Neurology 2001; 57: 1756-63. [PubMed]

68 Braak H, Braak E. Neuropathological staging of Alzheimer-related changes. Acta Neuropathol 1991; 82: 239-59. [PubMed]

69 Fox NC, Crum WR, Scahill RI, Stevens JM, Janssen JC, Rossor MN. Imaging of onset and progression of Alzheimer's disease with voxel-compression mapping of serial magnetic resonance images. Lancet 2001; 358: 201-05. [PubMed]

70 Woermann FG, Free SL, Koepp MJ, Ashburner J, Duncan JS. Voxel-by-voxel comparison of automatically segmented cerebral gray matter: A rater-independent comparison of structural MRI in patients with epilepsy. Neuroimage 1999; 10: 373-84. [PubMed]

71 Siegle GJ, Steinhauer SR, Thase ME, Stenger VA, Carter CS. Can't shake that feeling: event-related fMRI assessment of sustained amygdala activity in response to emotional information in depressed individuals. Biol Psychiatry 2002; 51: 693-707. [PubMed]

72 Wagner AD. Early detection of Alzheimer's disease: an fMRI marker for people at risk?. Nat Neurosci 2000; 3: 973-74. [PubMed]

73 Capizzano AA, Schuff N, Amend DL, et al. Subcortical ischemic vascular dementia: assessment with quantitative MR imaging and 1H MR spectroscopy. AJNR Am J Neuroradiol 2000; 21: 621-30. [PubMed]

74 Hsu YY, Du AT, Schuff N, Weiner MW. Magnetic resonance imaging and magnetic resonance spectroscopy in dementias. J Geriatr Psychiatry Neurol 2001; 14: 145-66. [PubMed]

a JA is at the The Wellcome Department of Imaging Neuroscience, Institute of Neurology, London, UK. b JGC is at the Department of Psychiatry, Washington University School of Medicine, St Louis, MO, USA. c CD is at the Section of Biomedical Image Analysis, Department of Radiology, University of Pennsylvania, Philadelphia, PA, USA. d NCF is at the Dementia Research Group, Institute of Neurology, University College London, London, UK. e GBF is at the Laboratory of Epidemiology and Neuroimaging, IRCCS San Giovanni di Dio–FBF, Brescia, Italy. f PMT is at the Laboratory of Neuroimaging, Department of Neurology, Division of Brain Mapping, University of California School of Medicine, Los Angeles, CA, USA.

 Correspondence: Dr Giovanni B Frisoni, Laboratory of Epidemiology and Neuroimaging, IRCCS San Giovanni di Dio–FBF, via Pilastroni 4, 25125 Brescia, Italy. Tel +39 0 30 3501361; fax +39 0 27 00435727 
Previous Next


The Journal

Info for Authors