## Abstract

Alzheimer's disease (AD) is characterized by a heterogeneous distribution of pathological changes throughout the brain. Magnetic resonance imaging can be used to investigate the regional distribution of cortical atrophy in AD in vivo. One marker for the disease-specific atrophy is the thickness of the cortical mantle across the brain, obtained with automated 3-D image processing. Here, we present data from 36 subjects (17 controls and, 19 patients diagnosed as probable AD) investigated for cortical thickness across the entire brain. We show significant cortical thickness decline in AD in temporal, orbitofrontal and parietal regions, with the most pronounced changes occurring in the allocortical region of the medial temporal lobes, outlining the parahippocampal gyrus, and representing a loss of >1.25 millimeters of cortical thickness. Moreover, focal cortical areas decline with progression of the disease as measured by time from baseline scan as well as the Mini-Mental State Exam. The results demonstrate the ability of this method to detect changes in cortical thickness in AD, across the entire brain, without need of prior anatomical definitions. The regional distribution of changes reported here is consistent with independent findings on the distribution of neuropathological alterations in AD. Using cortical thickness, moreover, we provide a direct quantitative index of atrophy in the disease.

## Introduction

Alzheimer's disease is characterized by the formation of neurofibrillary plaques and tangles, and neuronal loss across the central nervous system (Small et al., 2002). The histopathological changes show a characteristic sequence, with the entorhinal cortex and the hippocampus being among the first affected regions of the brain, followed by selected regions of the neocortex (Braak and Braak, 1991, 1996; Juottonen et al., 1998). Cognitive decline correlates with cortical atrophy (Mungas et al., 2002), which can be investigated with magnetic resonance imaging (MRI) in vivo.

MRI provides insight into the temporal sequence of AD-related regional atrophy, using region specific as well as global search algorithms. Within region specific protocols, recent effort has focused on using MRI to detect early morphological changes in AD pathology; attention has specifically focused on developing precise segmentation protocols for the early affected temporal lobe structures such as the entorhinal cortex (EC) or the hippocampus (HC) (Van Hoesen, 1995; de Leon et al., 1997; Mori et al., 1997; Krasuski et al., 1998; De Toledo-Morrell et al., 2000).

In comparison, whole brain imaging analysis allows detection of changes throughout the entire cerebrum. So far, the methodology of whole brain analysis approaches has been restricted to voxel-based morphometry (VBM) (Ashburner and Friston, 2000), implemented volumetrically (Baron et al., 2001) across the cortex (Thompson, 1989; Thompson et al., 2001), or as deformation analysis in longitudinal datasets (Good et al., 2002). In VBM, the concepts of gray matter density and gray matter concentration are central for the interpretation of the results; however, voxel density at any one point for any one subject is, unlike cortical thickness, meaningless.

Using specific algorithms to analyze cortical thickness across the entire cortex is a complementary method to the established research paradigms, offering a direct quantitative index of cortical atrophy that can be applied to single subjects and to group analysis. Cortical atrophy is reflected in a loss of gray matter which will result in a reduction of cortical thickness. Measuring cortical thickness across the entire cerebrum establishes a marker for the AD-related cortical atrophy.

## Materials and Methods

Thirty-six subjects were investigated. MRI scans were acquired from, 19 patients with a combined 31 acquisitions (up to three scans per subject). The patients had the clinical diagnosis of probable AD according to the NINCDS-ADRDA (McKhann et al., 1984). For comparison of baseline MRI measures, 17 healthy volunteers with one acquisition each were recruited and subsequently scanned using identical acquisition parameters. Patients were recruited from the Department of Psychiatry, Alzheimer Memorial Center, Dementia and Imaging Research Group, University of Munich, Germany. Further sociodemographic information of the subjects is shown in Table 1. Cognitive impairment in the AD patients was assessed using the Mini Mental State Examination (MMSE) (Folstein et al., 1975). The average MMSE score in the Alzheimer group was 21.2 (10–29). In the control group, the MMSE average score was 29.3 (range 28–30).

Table 1

Demographics of the study population

No. of subjects

No. of scans

Age

MMSE

Controls 17 17 61.0 ± 9.1 29.3 ± 0.6
Patients

19

31

68.8 ± 6.9

21.2 ± 4.6

No. of subjects

No. of scans

Age

MMSE

Controls 17 17 61.0 ± 9.1 29.3 ± 0.6
Patients

19

31

68.8 ± 6.9

21.2 ± 4.6

Significant medical co-morbidity in the AD patients and controls was excluded by interviews on medical history, physical and neurological examination, psychiatric evaluation, chest X-ray, ECG, EEG, brain MRI and laboratory tests (complete blood count, sedimentation rate, electrolytes, glucose, blood urea nitrogen, creatinine, liver-associated enzymes, cholesterol, HDL, triglycerides, antinuclear antibodies, rheumatoid factor, VDRL, HIV, serum B12, folate, thyroid function tests and urine analysis). None of the AD patients had hypertension or diabetes. All subjects or the holders of their Durable Power of Attorney provided written informed consent for the study. The protocol was approved by the Ethical Review Board of the Faculty of Medicine, Ludwig Maximilian University, Munich, Germany.

MRI examinations were performed on a 1.5 T Siemens Magnetom Vision MRI scanner (Siemens, Erlangen, Germany). All subjects were investigated with a volumetric T1-weighted sagittal oriented MRI sequence (TR = 11.6 ms, TE = 4.9 ms, resolution = 0.94 × 0.94 × 1.2 mm). The rectangular field of view (FOV) for the sagittal images was 256 mm (SI) × 204 mm (AP). Additionally, an axial-oriented fast FLAIR sequence (TR = 9000 ms, TE = 110 ms, resolution = 0.94 × 0.94 × 6 mm) was obtained. For the purpose of this study, only the T1-weighted images entered further processing. The native MRI were registered into standardized stereotaxic space using a linear transformation (Collins et al., 1994). Simultaneously, the images were corrected for non-uniformity artifacts (Sled et al., 1998). The registered and corrected volumes were segmented into white matter, gray matter, cerebrospinal fluid and background using an advanced neural net classifier (Zijdenbos et al., 2002). The white and gray matter surface were then fitted using deformable models (MacDonald et al., 2000), resulting in two surfaces with 81 920 polygons each. The surface deformation algorithm works by first fitting the white matter surface, then expanding outward to find the gray matter CSF intersection. One characteristic of this procedure is that each vertex of the white matter surface is closely related to its gray matter surface counterpart; cortical thickness can thus be defined as the distance between these linked vertices. The relevant parts of the processing pipeline are shown schematically in Figure 1.

Figure 1.

An overview of the steps involved in cortical thickness analysis. First, the images are non-uniformity corrected and registered into stereotaxic space. They are then classified (1) and fit with a white matter surface (2). The gray surface is found by expanding out from the white (3). Cortical thickness is measured at every vertex (4), and blurred using a 20 mm surface-based kernel (5).

Figure 1.

An overview of the steps involved in cortical thickness analysis. First, the images are non-uniformity corrected and registered into stereotaxic space. They are then classified (1) and fit with a white matter surface (2). The gray surface is found by expanding out from the white (3). Cortical thickness is measured at every vertex (4), and blurred using a 20 mm surface-based kernel (5).

In order to improve the ability to detect population changes, each subject's cortical thickness map was blurred using a 20 mm surface based blurring kernel (Chung et al., 2002). Diffusion smoothing, unlike the volumetric blurring used in VBM, follows the curvature of the surface and thus respects anatomical boundaries. Twenty millimeters was chosen as the kernel size in order to maximize statistical power while still minimizing false positives.

Statistical analysis was performed at every vertex, regressing cortical thickness against clinical state, MMSE scores, or time from baseline. Multiple time-points were used where available in order to provide greater stability in the estimation of the fixed effects than a purely cross-sectional model would have allowed. Linear mixed models using the restricted maximum likelihood (REML) estimation method were employed to account for the within subject correlations present due to the repeated acquisitions in the AD cohort (Pinheiro and Bates, 2000). Mixed models extend linear models by incorporating random effects, which can best be regarded as an additional error term. The model used is the following:

$Y_{i}{=}X_{i}\mathrm{{\beta}}{+}Z_{i}b_{i}{+}\mathrm{{\varepsilon}}_{i},{\,}i{=}1,{\ldots},M$
where β is the vector of fixed effects, bi, i = 1,…,M is the vector of random effects describing a shift in the intercept for each subject (M taking the value of either 19 or 36, depending on whether only the patients or all subjects were used), Xi (of size ni × p) and Z (of size ni × q) are known fixed-effects and random-effects regressor matrices and εi is the ni dimensional within group error vector, where ni takes on values between 1 and 3, describing the number of acquisitions per subject. yi is the estimated cortical thickness vector for subject i. Each of the statistical models used age as a covariate, thereby controlling for the main effects of age on the dependent variables; the value of p is therefore 3. The random-effects matrices (Z) included the intercept term; the value of q is therefore 1. For the regression with the MMSE scores, only the AD subjects were selected in order to avoid confounding of the clinical state with the MMSE analysis.

To illustrate with an example, the matrices for the MMSE analysis would be:

$X_{1}{=}{\cdots}{=}X_{19}{=}\left[\begin{array}{lll}1&66.9&24\\1&67.8&20\\1&69.5&20\end{array}\right],{\ }Z_{1}{=}{\cdots}{=}Z_{19}{=}\left[\begin{array}{l}1\\1\\1\end{array}\right]$
with the example subject 19 being one subject with three acquisitions taken at age 66.9, 67.8 and 69.5, and having MMSE scores of 24, 20 and 20 at each respective acquisition. The one-dimensional random effects matrix Z is the intercept for each subject — i.e. a common slope across subjects was assumed, with only the intercepts allowed to vary. For the other analyses (clinical state and time from baseline) the third column of the matrix would change to be a 0 or 1 for control or patient in the clinical state analysis (i.e. ‘treatment’ contrasts were used) or the time, measured in years, of the follow up scan minus the baseline scan (i.e. for subject 19, these would be 0, 0.9 and 2.6). The terms of β would thus be: β1, the mean intercept; β2, for the common effect of age; and β3, the common effect of MMSE/group/time from baseline. The one-dimensional vector bi, i = 1,…,19 (in the case of the MMSE or time since baseline analysis, since only the 19 patients are considered) or bi, i = 1,…,36 (in the case of the clinical state analysis, since all 36 subjects are considered) describes a shift in the intercept for each subject. In the parlance of the computational system used to solve these equations (Pinheiro and Bates, 2000), the function call is lme (y1 + age + MMSE, random= ∼1|ID), where y is the cortical thickness at the vertex in question, age is the age at the time of each acquisition, MMSE is the MMSE score at the time of each acquisition, the random effect is the intercept and all subjects are grouped by the subject ID.

The resulting statistical maps were thresholded using the false discovery rate (FDR) theory (Genovese et al., 2002) at a q value of 0.05 after pooling the P-values from all regressions run in this analysis. The interpretation of the maps is therefore that, on average, 5% of the results shown across all regressions are false positives. The figures also show the regression slope, which is the change in millimeters with each unit of measurement (group difference, point of MMSE, or in years).

Cortical thickness methods applied to human magnetic resonance images can be divided into three broad categories. The first measures thickness on a voxel by voxel level, and is illustrated by two papers solving a partial differential equation across the cortex (Jones et al., 2000; Yezzi and Prince, 2003). The second set of methods use advanced versions of the marching cubes algorithm, creating two cortical surfaces with variable tesselation (cf. Zeng et al., 1999; Miller et al., 2000). The last group uses deformable models to create white and grey matter cortices, and includes our work outlined above and that of Fischl and colleagues (Fischl and Dale, 2000; Rosas et al., 2002). One characteristic of using deformable models is that the number of polygons will always be identical across subjects, allowing for the easy creation of a surface coordinate system that can be used to run statistical analyses of cortical thickness at every vertex of the surface.

Our cortical thickness algorithm has been validated by comparison to manual measurements (Kabani et al., 2001). Furthermore, extensive analysis of the precision of our cortical thickness analysis framework was recently preformed (Lerch and Evans, 2004). Two tests were run: a study of reliability, where 19 scans of the same subject were processed separately to assess for methodological variability, and a population simulation. In the latter, 50 normal scans were processed. Twenty-five of them had the right superior temporal gyrus (rSTG) artificially thinned by one six-neighbor dilation of the white matter in that region, and that change recovered statistically. In the 19 repeat scans the mean standard deviation across all vertices was 0.27 mm. The population simulation resulted in 93% sensitivity in areas of perfect overlap of the rSTG along with 100% specificity (no single vertex outside of the rSTG was found to be significant between the two groups). This simulation study provides confidence that results shown using the AD data described in this paper are valid.

## Results

### Group Differences (Normals versus AD)

The results from the analysis clearly show significant differences in cortical thickness between the two groups. The average thickness across the entire cortex was significantly thinner in AD patients (3.1 ± 0.28 mm) compared with controls (3.74 ± 0.32 mm; t = −3.8, P < 0.0007), resulting in an average difference of 0.47 mm after removal of age effect by regression. Furthermore, the resulting maps of atrophy clearly show region specificity of thickness decline in AD. The most significant changes were found in the medial temporal lobes, the anterior and posterior cingulate region, the frontal lobes, the inferior parietal lobes, the orbitofrontal cortex, and the visual association cortex (Figs 2–4, Table 2). Most of these effects were found bilaterally except for the insula region, where only the left hemisphere appeared significant.

Figure 2.

t-statistical (lower panel) and cortical thickness (upper panel) difference maps in 19 AD subjects versus 17 controls (age effect removed by regression). Results from the statistical analysis are displayed at each vertex of the surface of a standardized brain in terms of t-statistical color maps as well as color maps displaying the estimated cortical thickness difference in millimeters (the regression slope) between the two groups. Significant differences can be seen in the temporal lobes, especially the entorhinal and perirhinal cortices, as well as medial frontal and parietal lobes and left associative visual areas.

Figure 2.

t-statistical (lower panel) and cortical thickness (upper panel) difference maps in 19 AD subjects versus 17 controls (age effect removed by regression). Results from the statistical analysis are displayed at each vertex of the surface of a standardized brain in terms of t-statistical color maps as well as color maps displaying the estimated cortical thickness difference in millimeters (the regression slope) between the two groups. Significant differences can be seen in the temporal lobes, especially the entorhinal and perirhinal cortices, as well as medial frontal and parietal lobes and left associative visual areas.

Figure 3.

t-statistical (lower panel) and cortical thickness (upper panel) difference maps in 19 AD subjects versus 17 controls (age effect removed by regression). The display emphasizes the differences in the left lateral and right medial views. Significant differences can be seen in the posterior and anterior cingulate, the left dorsolateral prefrontal cortex, most of the temporal lobes, and the left supramarginal gyrus.

Figure 3.

t-statistical (lower panel) and cortical thickness (upper panel) difference maps in 19 AD subjects versus 17 controls (age effect removed by regression). The display emphasizes the differences in the left lateral and right medial views. Significant differences can be seen in the posterior and anterior cingulate, the left dorsolateral prefrontal cortex, most of the temporal lobes, and the left supramarginal gyrus.

Figure 4.

t-statistical (lower panel) and cortical thickness (upper panel) difference maps in 19 AD subjects versus 17 controls (age effect removed by regression). The display emphasizes the differences right lateral and left medial views. Significant differences can be found throughout the temporal lobes, the posterior and anterior cingulate. Compared with the left hemisphere, the difference in the supramarginal gyrus and the dorsolateral prefrontal cortex is reduced.

Figure 4.

t-statistical (lower panel) and cortical thickness (upper panel) difference maps in 19 AD subjects versus 17 controls (age effect removed by regression). The display emphasizes the differences right lateral and left medial views. Significant differences can be found throughout the temporal lobes, the posterior and anterior cingulate. Compared with the left hemisphere, the difference in the supramarginal gyrus and the dorsolateral prefrontal cortex is reduced.

Table 2

Results sorted by t-statistics

Location Talairach coordinates

t-statistics Difference (mm) % atrophy

x

y

z

PHG 27 −39 −6.7 −1.25 ± 0.22 22 ± 3.8
ITG −47 −55 −16 −5.3 −1.00 ± 0.18 22 ± 4.0
MOG −13 31 −21 −4.5 −0.86 ± 0.23 18 ± 4.9
STG 59 −42 14 −4.4 −0.65 ± 0.14 17 ± 2.7
Post. Cing. −30 37 −4.3 −0.86 ± 0.19 22 ± 5.0
LOG −20 −99 −4.1 −1.03 ± 0.25 24 ± 5.8
MTG 63 −39 −14 −3.8 −0.85 ± 0.22 20 ± 5.2
Lt. IFG −46 36 11 −3.7 −0.76 ± 0.22 22 ± 5.6
Ant. cing.

−11

50

5

−3.4

−0.95 ± 0.28

19 ± 5.8

Location Talairach coordinates

t-statistics Difference (mm) % atrophy

x

y

z

PHG 27 −39 −6.7 −1.25 ± 0.22 22 ± 3.8
ITG −47 −55 −16 −5.3 −1.00 ± 0.18 22 ± 4.0
MOG −13 31 −21 −4.5 −0.86 ± 0.23 18 ± 4.9
STG 59 −42 14 −4.4 −0.65 ± 0.14 17 ± 2.7
Post. Cing. −30 37 −4.3 −0.86 ± 0.19 22 ± 5.0
LOG −20 −99 −4.1 −1.03 ± 0.25 24 ± 5.8
MTG 63 −39 −14 −3.8 −0.85 ± 0.22 20 ± 5.2
Lt. IFG −46 36 11 −3.7 −0.76 ± 0.22 22 ± 5.6
Ant. cing.

−11

50

5

−3.4

−0.95 ± 0.28

19 ± 5.8

Ant. cing., anterior cingulated; ITG, inferior tempral gyrus; LOG, lateral occipital gyrus; Lt IFG, left inferior frontal gyrus; MFG, middle frontal gyrus; MOG, medial orbital gyrus; MTG, middle temporal gyrus; PHG, parahippocampal gyrus; Post. cing., posterior cingulated; STG, superior temporal gyrus.

In the frontal lobes, the effects were most pronounced in the left anterior cingulate region (0.95 ± 0.28 mm loss in AD), the dorsolateral prefrontal cortex (0.76 ± 0.22 mm loss), and the orbitofrontal cortex (0.86 ± 0.23 mm). The effects in the dorsolateral prefrontal cortex appeared in the vicinity of Brodmann's area (BA) 45. The effects were stronger in the left than in the right hemisphere. In the parietal lobe, the effects were most pronounced in the posterior cingulate region (0.86 ± 0.19 mm loss in cortical thickness in AD), and the visual association areas (1.0 ± 0.25 mm loss).

The entire medial temporal lobe appeared to be severely affected. The average difference between cortical thickness in AD and in the control group ranged from 0.5 to 1.3 mm, with the strongest differences emerging in the area of the parahippocampal gyrus (PHG). This finding is consistent with the results from recent studies suggesting that the PHG, especially the entorhinal cortex (EC), is affected early in the course of AD (Krasuski et al., 1998; Van Hoesen et al., 2000; Callen et al., 2001). To investigate cortical thickness decline in this area in our subjects more closely, manual segmentation of the structures of the PHG was used to trace the EC in the MRI of all subjects using a recently developed protocol (Pruessner et al., 2002). The labels of the EC from all subjects were then used to create a customized probabilistic map of the EC for the subjects in this study, which was overlaid onto the t-statistics map from the thickness analysis. This allowed an accurate evaluation of the decline in cortical thickness within the EC in our study sample. The result of this procedure is shown in Figure 5, indicating that the most striking difference between the two groups occurred in the anterior portion of the EC in the left hemisphere, with a difference of 1.25 ± 0.22 mm in cortical thickness between the two groups. The lateral temporal lobes also showed significant interaction with clinical state, with a peak of 0.8 ± 0.22 mm in the left medial temporal gyrus (BA 21).

Figure 5.

Regional analysis using the cortical thickness methodology in AD versus normal controls displaying differences in the entorhinal cortex. The four cortical views show: (a) the probability maps of the entorhinal and perirhinal cortices in this population; (b) the t-statistics of the MMSE regression; (c) the group analysis; and (d) the time difference from baseline. The color scales are the same as in Figures 2–4, 6 and 7 respectively. The graphs illustrate the effects at the vertex indicated by the black lines. Atrophy of the PHG is clearly implicated in each of the three analyses shown.

Figure 5.

Regional analysis using the cortical thickness methodology in AD versus normal controls displaying differences in the entorhinal cortex. The four cortical views show: (a) the probability maps of the entorhinal and perirhinal cortices in this population; (b) the t-statistics of the MMSE regression; (c) the group analysis; and (d) the time difference from baseline. The color scales are the same as in Figures 2–4, 6 and 7 respectively. The graphs illustrate the effects at the vertex indicated by the black lines. Atrophy of the PHG is clearly implicated in each of the three analyses shown.

In the parietal and occipital lobes, significant differences further appeared in the transition area between parietal and occipital lobes in the area of the visual association cortex (BA 18–19; between 0.4 – 1.0 mm loss of cortical thickness in the AD group). Interestingly, the primary visual cortex (BA 17) appeared to be spared, showing no interaction with group (t = −1.6).

### Regression of Cortical Thickness versus MMSE

In order to investigate whether cortical thickness measures vary with disease progression in the AD patients, a regression of cortical thickness against the MMSE score was performed in the AD patients at every vertex of the surface model. Mean cortical thickness was marginally associated with MMSE scores (t = 2.24, P = 0.06). Regionally, however, it was found that lower MMSE scores were associated with significantly thinner cortical thickness in the bilateral PHG, the left superior temporal gyrus, left insula, and left anterior cingulate gyrus (Fig. 6).

Figure 6.

t-statistical (right column) and cortical thickness (left column) maps of the MMSE regression against cortical thickness in 19 AD subjects. Significant correlations can be seen in the lateral temporal lobes, the left insula, left anterior cingulate and bilateral para-hippocampal gyri.

Figure 6.

t-statistical (right column) and cortical thickness (left column) maps of the MMSE regression against cortical thickness in 19 AD subjects. Significant correlations can be seen in the lateral temporal lobes, the left insula, left anterior cingulate and bilateral para-hippocampal gyri.

### Regression of Cortical Thickness versus Progression

Further analysis of the effect of disease progression was performed by regressing cortical thickness against the time difference from baseline in the follow-up scans within the AD cohort. There was a significant correlation between time from baseline and thickness (t = −3.47, P = 0.006) featuring a decline of 0.18 mm per year. Significant regional correlations were found in the anterior temporal lobes including the para-hippocampal gyrus, the anterior frontal lobes, and the anterior cingulate (Fig. 7). A post-hoc test was also conducted to ascertain the rate of thinning in cortical areas that featured an initial group difference versus those where the Alzheimer's patients did not have significantly thinner cortex than the normal controls. Both these areas declined significantly with time from baseline; the rate was faster in cortical regions which were also significantly thinner in the patients (0.21mm per year, t = −4, P = 0.004) than the rest of the cortex (0.16mm per year, t = −3.4, P = 0.009).

Figure 7.

t-statistical (right column) and cortical thickness (left column) maps of the regression against time from baseline. Significant correlations can be seen in the anterior temporal lobes, anterior frontal lobes, and lateral posterior cortex.

Figure 7.

t-statistical (right column) and cortical thickness (left column) maps of the regression against time from baseline. Significant correlations can be seen in the anterior temporal lobes, anterior frontal lobes, and lateral posterior cortex.

## Discussion

We used a fully automated method to measure cortical thickness across the entire brain to investigate differences in a group of AD patients versus age-matched controls. We further investigated the correlation between cortical thickness and MMSE scores as well as time since baseline in the patient population.

Our results clearly show AD related decline in cortical thickness in multiple areas of the brain, many of which have been reported in previous MR studies. Cortical thickness of the medial temporal lobes was most severely reduced in AD patients, and, within the medial temporal lobes, the parahippocampal gyrus was most affected. This is consistent with previous MR and histopathological studies showing that this region of the brain is affected early and profoundly in the course of the disease (Juottonen et al., 1999; Xu et al., 2000). The results from this study extend previous findings by showing that in the left hemisphere, the posterior portions of the PHG also seem to be strongly affected in AD. For the anterior portion of the PHG, in the area of the entorhinal cortex, the most significant differences occurred in the left hemisphere, though the effects were largely bilateral.

Cortical atrophy in AD is not limited to the medial temporal lobe; the remaining limbic system, the lateral temporal lobes, and certain associative visual areas correlate significantly with the disease as well. The finding of significant cortical thickness decline in the area of the visual association cortex is consistent with other studies showing this region to be strongly affected by AD (Cronin-Golomb et al., 1991; Thulborn et al., 2000), as are the results in the lateral temporal lobes (Busatto et al., 2003). Within the limbic system the orbito-frontal cortex (Van Hoesen et al., 2000) and the anterior and posterior cingulate cortices (Valla et al., 2001; Matsuda et al., 2002) have also been shown to correlate with the disease in ways that support the results of this study. Combined with spared areas of the cortex, such as the motor, sensory and primary visual cortices (BA17) (Buckner et al., 2000; Rizzo et al., 2000; Prvulovic et al., 2002), a pattern of atrophy in AD emerges that is focal to multiple cortical areas while centered on the medial temporal lobes.

The results from the MMSE regression within the AD group indicate that the decline in the PHG is correlated with the MMSE score (Fields et al., 1992). Significant associations between MMSE scores and cortical thickness were found in bilateral medial temporal areas as well as temporal, anterior cingulate, and insular regions of the left hemisphere. The results demonstrate that cortical thickness analysis can be used to link clinical information with decline in particular cortical areas when available. In the case of MMSE scores, the significant findings with cortical atrophy were left hemisphere dominant. Progression as measured by time from baseline in repeated scans showed a more significant effect than the MMSE analysis, indicating continuing atrophy that is not entirely captured by changes in cognitive state as captured by the MMSE. Progressive atrophy was particularly strong in the anterior frontal and temporal lobes as well as the posterior cingulate. Cortical areas which were significantly thinner in the patients thinned at a faster rate than those that were not, though this difference was small and would therefore suggest that the entire cortex is declining with disease progression. The longitudinal aspect of our analysis should be explored in greater detail and compared to non-linear registration methods that have recently been used in the analysis of AD populations and normal aging (Fox et al., 2000; Scahill et al., 2002, 2003).

AD is commonly subdivided into six stages representing advancing pathology as defined by the progression of neurofibrillary tangles (NFTs) and senile placques (Braak and Braak, 1991, 1996). The patients used in this study were diagnosed with probable AD according to NINCDS-ARDRA criteria and thus believed to be in stage V or VI (Nagy et al., 1999). These two end stages, known as the neocortical stages, feature plaque and NFT involvement in virtually all subdivisions of the cerebral cortex with a particular emphasis on association areas and the medial temporal lobes. NFT presence is strongly correlated with neuronal loss and cortical atrophy (Gomez-Isla et al., 1997; Grignon et al., 1998). Moreover, increased duration of NFT presence is associated with increased atrophy (Grignon et al., 1998). This leads to the hypothesis that, in patients diagnosed with probable AD, cortical areas involved earlier in the disease progression will show greater atrophy than areas involved later (Gomez-Isla et al., 1997). Our results, which reveal greatest MRI detectable atrophy in areas implicated earlier (such as the entorhinal cortex), largely support this hypothesis. The ability to stage AD based on in-vivo imaging should be investigated further. Moreover, some of the findings in this study may be due to the small sample size, since occasional atypical representation of AD (such as the visuo-spatial variant; cf. Cronin-Golomb et al., 1991) may have an undue influence on our results. This analysis should therefore be replicated in a larger sample.

The major difference between previous findings and the results obtained in this study lies in the use of the cortical thickness analysis technique. This strategy has a number of advantages when compared with other MR methods used to investigate AD with MRI. Cortical thickness analysis allows searching for associations between the depth of the cortex and sociodemographic, clinical or psychological variables across the entire surface of the brain. It shares this advantage with other global search algorithms like VBM and deformation field analysis. Unlike VBM, however, it is based on the description of the actual thickness of the cortex in millimeters and thus allows meaningful quantitative description of the results (difference in cortical thickness in millimeters). The major disadvantage is that the focus is entirely on the cortex; changes in subcortical structures, white matter or cerebro-spinal fluid will not be picked up.

Finally, cortical thickness analysis is fully automated and thus reaches 100% operator independence. The results are operator-independent and do not rely on the correct interpretation of regional anatomical variations, in contrast to manual segmentation protocols.

Cortical thickness analysis may serve as a surrogate marker for the neuronal loss that accompanies the histopathological changes in the cortex which occur in AD and their temporal ordering. For example, the highly significant findings in the PHG, in the area of the EC, support the view (Juottonen et al., 1999; Xu et al., 2000) about the importance of medial temporal lobe morphometry in the diagnosis of AD. Future studies can compare the relative efficacy of cortical thickness measures with volumetric analyses of the hippocampus (Pruessner et al., 2001).

Continuing work will investigate associations between cortical thickness and other sociodemographic or clinical variables. Owing to its fully automatic implementation, the method can be used to characterize the healthy and pathological ranges of cortical thickness in specific age groups. The development of normative data for specific age and disease groups will allow direct comparison of individual subjects with cortical thickness norms in health and disease. This process has the potential of aiding in the early diagnosis of dementia. Furthermore, unlike commonly used region of interest measures, cortical thickness analysis provides coverage of the complete cerebrum, and could thus be used for differential diagnosis of the various types of dementia as well. Finally, future studies will have to show the value of this new method in monitoring of the progression of dementia across the entire cortex.

ICBM grant PO1MHO52176-11, principal investigator Dr John Mazziotta; CIHR grant MOP-34996. Jason Lerch is funded by a K.M. Hunter/CIHR Doctoral Research Award.

## References

Ashburner J, Friston KJ (
2000
) Voxel-based morphometry — the methods.
Neuroimage

11
:
805
–821.
Baron JC, Chetelat G, Desgranges B, Perchey G, Landeau B, de la Sayette V, Eustache F (
2001
) In vivo mapping of gray matter loss with voxel-based morphometry in mild Alzheimer's disease.
Neuroimage

14
:
298
–309.
Braak H, Braak E (
1991
) Neuropathological stageing of Alzheimer-related changes.
Acta Neuropathol (Berl)

82
:
239
–259.
Braak H, Braak E (
1996
) Evolution of the neuropathology of Alzheimer's disease.
Acta Neurol Scand Suppl

165
:
3
–12.
Buckner RL, Snyder AZ, Sanders AL, Raichle ME, Morris JC (
2000
) Functional brain imaging of young, nondemented, and demented older adults.
J Cogn Neurosci

12
(Suppl 2):
24
–34.
Busatto GF, Garrido GE, Almeida OP, Castro CC, Camargo CH, Cid CG, Buchpiguel CA, Furuie S, Bottino CM (
2003
) A voxel-based morphometry study of temporal lobe gray matter reductions in Alzheimer's disease.
Neurobiol Aging

24
:
221
–231.
Callen DJ, Black SE, Gao F, Caldwell CB, Szalai JP (
2001
) Beyond the hippocampus: MRI volumetry confirms widespread limbic atrophy in AD.
Neurology

57
:
1669
–1674.
Chung M, Worsley K, Paus T, Robbins S, Evanc A, Taylor J, Giedd J, Rapoport J (
2002
) Tensor-based surface morphometry. Madison, WI: University of Wisconsin.
Collins DL, Neelin P, Peters TM, Evans AC (
1994
) Automatic 3-D intersubject registration of MR volumetric data in standardized Talairach space.
J Comput Assist Tomogr

18
:
192
–205.
Cronin-Golomb A, Corkin S, Rizzo JF, Cohen J, Growdon JH, Banks KS (
1991
) Visual dysfunction in Alzheimer's disease:relation to normal aging.
Ann Neurol

29
:
41
–52.
de Leon MJ, Convit A, DeSanti S, Bobinski M, George AE, Wisniewski HM, Rusinek H, Carroll R, Saint Louis LA (
1997
) Contribution of structural neuroimaging to the early diagnosis of Alzheimer's disease.
Int Psychogeriatr

9
(Suppl 1):
183
–90 [discussion 247–52].
De Toledo-Morrell L, Goncharova I, Dickerson B, Wilson RS, Bennett DA (
2000
) From healthy aging to early Alzheimer's disease: in vivo detection of entorhinal cortex atrophy.

911
:
240
–253.
Fields SD, Fulop G, Sachs CJ, Strain J, Fillit H (
1992
) Usefulness of the Neurobehavioral Cognitive Status Examination in the hospitalized elderly.
Int Psychogeriatr

4
:
93
–102.
Fischl B, Dale AM (
2000
) Measuring the thickness of the human cerebral cortex from magnetic resonance images.

97
:
11050
–11055.
Folstein MF, Folstein SE, McHugh PR (
1975
) ‘Mini-mental state’. A practical method for grading the cognitive state of patients for the clinician.
J Psychiatr Res

12
:
189
–198.
Fox NC, Cousens S, Scahill R, Harvey RJ, Rossor MN (
2000
) 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

57
:
339
–344.
Genovese CR, Lazar NA, Nichols T (
2002
) Thresholding of statistical maps in functional neuroimaging using the false discovery rate.
Neuroimage

15
:
870
–878.
Gomez-Isla T, Hollister R, West H, Mui S, Growdon JH, Petersen RC, Parisi JE, Hyman BT (
1997
) Neuronal loss correlates with but exceeds neurofibrillary tangles in Alzheimer's disease.
Ann Neurol

41
:
17
–24.
Good CD, Scahill RI, Fox NC, Ashburner J, Friston KJ, Chan D, Crum WR, Rossor MN, Frackowiak RS (
2002
) Automatic differentiation of anatomical patterns in the human brain: validation with studies of degenerative dementias.
Neuroimage

17
:
29
–46.
Grignon Y, Duyckaerts C, Bennecib M, Hauw JJ (
1998
) Cytoarchitectonic alterations in the supramarginal gyrus of late onset Alzheimer's disease.
Acta Neuropathol (Berl)

95
:
395
–406.
Jones SE, Buchbinder BR, Aharon I (
2000
) Three-dimensional mapping of cortical thickness using Laplace's equation.
Hum Brain Mapp

11
:
12
–32.
Juottonen K, Laakso MP, Insausti R, Lehtovirta M, Pitkanen A, Partanen K, Soininen H (
1998
) Volumes of the entorhinal and perirhinal cortices in Alzheimer's disease.
Neurobiol Aging

19
:
15
–22.
Juottonen K, Laakso MP, Partanen K, Soininen H (
1999
) Comparative MR analysis of the entorhinal cortex and hippocampus in diagnosing Alzheimer disease.

20
(1):
139
–44.
Kabani N, Le Goualher G, MacDonald D, Evans AC (
2001
) Measurement of cortical thickness using an automated 3-D algorithm: a validation study.
Neuroimage

13
:
375
–380.
Krasuski JS, Alexander GE, Horwitz B, Daly EM, Murphy DG, Rapoport SI, Schapiro MB (
1998
) Volumes of medial temporal lobe structures in patients with Alzheimer's disease and mild cognitive impairment (and in healthy controls).
Biol Psychiatry

43
:
60
–68.
Lerch JP, Evans AC (
2004
) Cortical thickness analysis examined through power analysis and a population simulation. Neuroimage (in press).
MacDonald D, Kabani N, Avis D, Evans AC (
2000
) Automated 3-D extraction of inner and outer surfaces of cerebral cortex from MRI.
Neuroimage

12
:
340
–356.
Matsuda H, Kitayama N, Ohnishi T, Asada T, Nakano S, Sakamoto S, Imabayashi E, Katoh A (
2002
) Longitudinal evaluation of both morphologic and functional changes in the same individuals with Alzheimer's disease.
J Nucl Med

43
:
304
–311.
McKhann G, Drachman D, Folstein M, Katzman R, Price D, Stadlan EM (
1984
) Clinical diagnosis of Alzheimer's disease:report of the NINCDS-ADRDA Work Group under the auspices of Department of Health and Human Services Task Force on Alzheimer's Disease.
Neurology

34
:
939
–944.
Miller MI, Massie AB, Ratnanather JT, Botteron KN, Csernansky JG (
2000
) Bayesian construction of geometrically based cortical thickness metrics.
Neuroimage

12
:
676
–687.
Mori E, Yoneda Y, Yamashita H, Hirono N, Ikeda M, Yamadori A (
1997
) Medial temporal structures relate to memory impairment in Alzheimer's disease:an MRI volumetric study.
J Neurol Neurosurg Psychiatry

63
:
214
–221.
Mungas D, Reed BR, Jagust WJ, DeCarli C, Mack WJ, Kramer JH, Weiner MW, Schuff N, Chui HC (
2002
) Volumetric MRI predicts rate of cognitive decline related to AD and cerebrovascular disease.
Neurology

59
:
867
–873.
Nagy Z, Hindley NJ, Braak H, Braak E, Yilmazer-Hanke DM, Schultz C, Barnetson L, Jobst KA, Smith AD (
1999
) Relationship between clinical and radiological diagnostic criteria for Alzheimer's disease and the extent of neuropathology as reflected by ‘stages’: a prospective study.
Dement Geriatr Cogn Disord

10
:
109
–114.
Pinheiro JC, Bates DM (
2000
) Mixed-effects models in S and S-PLUS. New York: Springer.
Pruessner JC, Collins DL, Pruessner M, Evans AC (
2001
) Age and gender predict volume decline in the anterior and posterior hippocampus in early adulthood.
J Neurosci

21
:
194
–200.
Pruessner JC, Kohler S, Crane J, Pruessner M, Lord C, Byrne A, Kabani N, Collins DL, Evans AC (
2002
) Volumetry of temporopolar, perirhinal, entorhinal and parahippocampal cortex from high-resolution MR images: considering the variability of the collateral sulcus.
Cereb Cortex

12
:
1342
–1353.
Prvulovic D, Hubl D, Sack AT, Melillo L, Maurer K, Frolich L, Lanfermann H, Zanella FE, Goebel R, Linden DE, Dierks T (
2002
) Functional imaging of visuospatial processing in Alzheimer's disease.
Neuroimage

17
:
1403
–1414.
Rizzo M, Anderson SW, Dawson J, Nawrot M (
2000
) Vision and cognition in Alzheimer's disease.
Neuropsychologia

38
:
1157
–1169.
Rosas HD, Liu AK, Hersch S, Glessner M, Ferrante RJ, Salat DH, van der Kouwe A, Jenkins BG, Dale AM, Fischl B (
2002
) Regional and progressive thinning of the cortical ribbon in Huntington's disease.
Neurology

58
:
695
–701.
Scahill RI, Schott JM, Stevens JM, Rossor MN, Fox NC (
2002
) Mapping the evolution of regional atrophy in Alzheimer's disease:unbiased analysis of fluid-registered serial MRI.

99
:
4703
–4707.
Scahill RI, Frost C, Jenkins R, Whitwell JL, Rossor MN, Fox NC (
2003
) A longitudinal study of brain volume changes in normal aging using serial registered magnetic resonance imaging.
Arch Neurol

60
:
989
–994.
Sled JG, Zijdenbos AP, Evans AC (
1998
) A nonparametric method for automatic correction of intensity nonuniformity in MRI data.
IEEE Trans Med Imaging

17
:
87
–97.
Small GW, Agdeppa ED, Kepe V, Satyamurthy N, Huang SC, Barrio JR (
2002
) In vivo brain imaging of tangle burden in humans.
J Mol Neurosci

19
:
323
–327.
Thompson KM (
1989
) Effects of early alcohol use on adolescents' relations with peers and self-esteem: patterns over time.

24
:
837
–849.
Thompson PM, Mega MS, Woods RP, Zoumalan CI, Lindshield CJ, Blanton RE, Moussai J, Holmes CJ, Cummings JL, Toga AW (
2001
) Cortical change in Alzheimer's disease detected with a disease-specific population-based brain atlas.
Cereb Cortex

11
:
1
–16.
Thulborn KR, Martin C, Voyvodic JT (
2000
) Functional MR imaging using a visually guided saccade paradigm for comparing activation patterns in patients with probable Alzheimer's disease and in cognitively able elderly volunteers.

21
:
524
–531.
Valla J, Berndt JD, Gonzalez-Lima F (
2001
) Energy hypometabolism in posterior cingulate cortex of Alzheimer's patients:superficial laminar cytochrome oxidase associated with disease duration.
J Neurosci

21
:
4923
–4930.
Van Hoesen GW (
1995
) Anatomy of the medial temporal lobe.
Magn Reson Imag

13
:
1047
–1055.
Van Hoesen GW, Parvizi J, Chu CC (
2000
) Orbitofrontal cortex pathology in Alzheimer's disease.
Cereb Cortex

10
:
243
–251.
Xu Y, Jack CR Jr, O'Brien PC, Kokmen E, Smith GE, Ivnik RJ, Boeve BF, Tangalos RG, Petersen RC (
2000
) Usefulness of MRI measures of entorhinal cortex versus hippocampus in AD.
Neurology

54
:
1760
–1767.
Yezzi AJ Jr, Prince JL (
2003
) An Eulerian PDE approach for computing tissue thickness.
IEEE Trans Med Imaging

22
:
1332
–1339.
Zeng X, Staib LH, Schultz RT, Duncan JS (
1999
) Segmentation and measurement of the cortex from 3-D MR images using coupled-surfaces propagation.
IEEE Trans Med Imag

18
:
927
–937.
Zijdenbos AP, Forghani R, Evans AC (
2002
) Automatic ‘pipeline’ analysis of 3-D MRI data for clinical trials: application to multiple sclerosis.
IEEE Trans Med Imag

21
:
1280
–1291.

## Author notes

1McConnell Brain Imaging Centre, Montreal Neurological Institute, McGill University, Montreal, Quebec, Canada, and 2Alzheimer Memorial Center, Dementia Research Section and Memory Clinic, Department of Psychiatry, Ludwig-Maximilian University, Munich, Germany