Imaging markers of mild cognitive impairment: Multivariate analysis of CBF SPECT

Neurobiology of Aging 28 (2007) 1062–1069

Imaging markers of mild cognitive impairment: Multivariate analysis of CBF SPECT Chaorui Huang a,b,∗ , David Eidelberg a , Christian Habeck c , James Moeller c,e , Leif Svensson d , Tyler Tarabula e , Per Julin b,f a

Center for Neurosciences, North Shore-Long Island Jewish Health System, New York University School of Medicine, New York, NY, USA b Karolinska Institute, Neurotec Department, Division of Clinical Geriatrics, Karolinska University Hospital, Sweden c Cognitive Neuroscience Division of the Taub Institute for Research in Alzheimer’s Disease and the Aging Brain, College of Physicians and Surgeons of Columbia University, New York, NY, USA d Karolinska Institute, Department of Hospital Physics, Karolinska University Hospital, Sweden e Department of Psychiatry, College of Physicians and Surgeons of Columbia University, New York, NY, USA f AstraZeneca R&D Neuroscience, Sodertalje, Sweden Received 9 February 2006; received in revised form 19 April 2006; accepted 3 May 2006 Available online 5 July 2006

Abstract This study aimed to investigate cross-sectional and longitudinal changes of regional cerebral blood flow (rCBF) in preclinical dementia using single photon emission computed tomography (SPECT). SPECT and cognitive function were investigated in 39 mild cognitive impairment (MCI) subjects and 20 age-matched controls. All subjects were followed longitudinally 19 months on average, 16 MCI subjects progressed to Alzheimer’s disease (AD), who were retrospectively defined as progressive mild cognitive impairment (PMCI) at baseline and 23 MCI subjects remained stable and were defined as stable mild cognitive impairment (SMCI) at baseline. SPECT was performed both at the initial investigation and at follow-up. Image data were analyzed using multivariate analysis, SPM and volume of interest (VOI)-based analysis. Significant covariate patterns were derived, which differentiate among PMCI, SMCI and controls at baseline as well as describe the longitudinal progression of PMCI. The combined SPECT and neuropsychology increased the diagnostic accuracy of PMCI at baseline. SPECT and neuropsychological testing can be used objectively for both baseline diagnosis and to monitor changes in brain function during very early AD. © 2006 Elsevier Inc. All rights reserved. Keywords: Single photon emission computed tomography (SPECT); Mild cognitive impairment (MCI); Alzheimer’s disease (AD); Neuronal network; Ordinal trends analysis (OrT); Cognitive function

1. Introduction Mild cognitive impairment (MCI) is conceptualized as a boundary state between aging and dementia [1]. The term refers to individuals whose memory or other cognitive abilities are not normal, but do not meet conventional criteria for ∗ Corresponding author at: Center for Neurosciences, Feinstein Institute for Medical Research, North Shore-Long Island Jewish Health System, 350 Community Drive, Manhasset, NY 11030, USA. Tel.: +1 516 562 1352; fax: +1 516 562 1008. E-mail address: [email protected] (C. Huang).

0197-4580/$ – see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.neurobiolaging.2006.05.017

dementia. Most of the MCI subjects will progress to dementia, especially Alzheimer’s disease (AD), at a rate of 10–15% per year, as compared to the normal aging who convert at a rate of 1–2% per year [15]. The current MCI research mainly concerns the early diagnosis of dementia and pharmacologic intervention. Thus, a thorough understanding of genesis and natural history of Alzheimer disease is necessary, which will lead to a number of potential clinical applications, including the diagnosis, treatment, and prevent of the onset and progression of the disease. Single photon emission computed tomography (SPECT) has been shown to be useful for the diagnosis of AD

C. Huang et al. / Neurobiology of Aging 28 (2007) 1062–1069

[9]. Reduced regional cerebral blood flow (rCBF) in parietal lobe and parieto-temporal association cortex are the typical findings in early AD. In MCI, SPECT has also been shown to be able to predict conversion to AD. A decreased rCBF in parietal lobe, posterior cingulate and precuni was found to be related to the future development of dementia [10,11,13]. Regarding to the longitudinal rCBF changes of MCI, selective rCBF reduction was observed in the left hippocampus and parahippocampus gyrus [13]. Concerning the neuropsychological evaluation, MCI had impaired cognitive function at baseline and declined significantly faster concerning episodic memory, semantic memory and perceptual speed during the follow-up, but working memory was spared [3]. In brain imaging data analysis, multivariate analysis techniques have recently received increasing attention. The technique is based upon principal component analysis (PCA), which evaluates correlation of activation across brain regions. Scaled subprofile model (SSM) analysis is one type of multivariate analyses and could be applied in the cross-sectional study to identify the covariate pathological brain networks in diseased groups. Ordinal trends (OrT) analysis is the method of choice in the analysis of neuroimaging data from experiments with parametric designs involving two or more task condition [6,8]. It focuses on neural processes for which the associated covariance patterns exhibit ordinal trends, i.e., the subject scores increase monotonically with changes in experimental conditions or at different time points. While originally developed for the study of time series fMRI data, this approach can also be used to model the specific network changes that occur in PET and SPECT studies of disease progression, as well as serial imaging studies of treatment effects and medication washout in the resting condition. The results of multivariate analysis can be more easily interpreted as a signature of neuronal networks. Univariate approaches, on the other hand, cannot directly address functional connectivity in the brain. The multivariate approach can also result in greater statistical power when compared with univariate techniques, which are forced to employ very stringent, and often overly conservative, corrections for voxel-wise multiple comparisons. Multivariate techniques also lend themselves much better to prospective application of results from the analysis of one dataset to entirely new datasets. Multivariate techniques are thus well placed to provide information about mean differences and correlations with behavior, similarly to univariate approaches, with potentially greater statistical power and better reproducibility checks. Based on this background, we aimed to examine the natural history of MCI using SPECT and neuropsychology and evaluated baseline differences and longitudinal progression of MCI. SSM analyses were applied in SPECT data among the groups of PMCI, SMCI and normal controls at the initial investigation, and OrT analysis was performed in PMCI and SMCI, both separately and combined, to model the disease progression. Neuropsychological tests were correlated with covariate pattern expression.

1063

2. Patients and methods 2.1. Subjects Thirty-nine MCI subjects and twenty controls were evaluated. The MCI subjects were selected from those individuals consecutively investigated for suspected dementia at the Geriatric Clinic, Karolinska University Hospital. The control subjects were recruited through advertisements in the press, the Swedish Pensioner Society and a Driving and Aging project. All subjects underwent SPECT and neuropsychological examination at the initial investigation. The MCI subjects were followed clinically for an averaged interval of 18.7 ± 8.7 months. The subjects underwent a second clinical evaluation as well as neuropsychological and SPECT examination. Sixteen MCI subjects progressed to AD. The MCI subjects were retrospectively diagnosed as progressive mild cognitive impairment (PMCI) at baseline. Twenty-three subjects remained stable, which were retrospectively diagnosed as stable mild cognitive impairment (SMCI) at baseline. The baseline values of PMCI, SMCI and controls did not differ at baseline with respect to age (PMCI: 61.6 ± 7.2 (years), SMCI: 58.7 ± 9.5 (years), controls: 61.3 ± 8.0 (years), p = 0.4853), gender (PMCI (f/m): 8/8, SMCI (f/m): 14/9, controls (f/m): 13/7, p = 0.6485) and follow-up time (PMCI: 18.9 ± 8.6 months, range: 9–41 months; SMCI: 18.5 ± 9.0 months, range: 9–39 months, p = 0.8741). No subjects received either psychotropic medication or an acetylcholinesterase inhibitor likely to influence the results of SPECT scanning. 2.2. Diagnosis Subjects who were diagnosed as MCI performed at least 1.5 S.D. below average for their age on at least one neuropsychological test, but did not fulfill the diagnostic criteria for dementia according to DSM-IV criteria [18] and did not have evidence of impairment in social or occupational functioning. Other medical conditions likely to explain the cognitive impairment were excluded during the clinical examination, which included a routine MRI scan. PMCI and SMCI are retrospective diagnostic terms based on the clinical follow-up. PMCI referred to the MCI subjects who converted to dementia according to the DSM-IV criteria during the follow-up. Whereas, SMCI was defined as the subjects who still did not fulfill the criteria for dementia according to DSM-IV during the observation time. 2.3. Neuropsychological tests All subjects were examined by an experienced psychologist in five cognitive domains using nine psychological tests. The five cognitive domains evaluated were episodic memory, semantic memory, visuospatial function, attention and general cognitive function. The nine psychological tests included

1064

C. Huang et al. / Neurobiology of Aging 28 (2007) 1062–1069

four subtests (information, similarities, block design and digit symbol) from the Wechsler adult intelligence scale-revised (WAIS-R) [17], trail making test (TMT) A and B, recognition words test from the Stockholm Geriatric Research Center (SGRC) [2] and Mini-Mental State Examination (MMSE) [5]. 2.4. Single photon emission computed tomography Each subject was injected with 1000 MBq Tc-99mHMPAO (Ceretec, Amersham Ltd.) in a quiet surrounding with eyes closed. Acquisition started 30 min after injection. Data were collected in 64 projections evenly spread through 360◦ with a single headed rotating gamma camera (Siemens Diacam) with a total acquisition time of 32 min. Tomographic slices were reconstructed using an iterative algorithm (Hosem, Nuclear Diagnostics AB, Sweden) with Chang attenuation correction (attenuation coefficient: 0.12 cm−1 ). Data were formatted as a 3D dataset with 64 × 64 × 64 cubic voxels with 3.5 mm sides. The resolution in a tomographic slice was measured to be 10.2 mm (FWHM). The reconstructed datasets were post-filtered with a Butterworth filter, cutoff 1.0 cm−1 . 2.5. SPECT registration and quantification 2.5.1. Imaging preprocessing All images were converted from the interfile 3.3 to the Analyzed format using a freeware medical image conversion program MedCon. The image preprocessing procedure was done using SPM99 (Welcome Department of Cognitive Neurology, London, UK) running on Matlab 6.0 (Mathworks Inc., Sherborn, MA). Image data were spatially normalized to a Talairach based template of rCBF using a 12 parameter affine transform, non-linear transformations and trilinear interpolation. The resultant voxel size was 2 mm × 2 mm × 2 mm. The normalized data were then smoothed using a Gaussian kernel at FWHM = 16 mm. 2.5.2. SSM analysis at baseline We conducted three voxel-based network analyses (software available at http://www.neuroscience-nslij.org) to identify the disease-related patterns of regional rCBF covariation in the combined groups of PMCI and SMCI, PMCI and controls as well as SMCI and controls separately. This data-driven principal components analysis (PCA) maps the regions contributing to the network onto standardized Talairach MRI sections without the need for a priori regionof-interest (ROI) placement. The details were described somewhere else [14]. We considered rCBF patterns resulting from the analyses to be disease-related if the subject scores (an index of network expression) discriminated the diseased group and reference group at p < 0.05. The expression scores were correlated with neuropsychological evaluation.

2.5.3. Ordinal trends analysis (OrT) for longitudinal data OrT is a general form of the PCA approach, which focuses on longitudinal changes in regional blood flow that keep the functional connectivity unchanged from the baseline condition and can therefore be captured in a single covariance pattern. In our case, this pattern represents a set of brain regions whose mutual interrelationships that are not changing from baseline to follow-up. The expression of the covariance pattern varies from subject to subject with the additional constraint of an increase from baseline for as many subjects as possible. This means that most patients will display the same mutually correlated regional metabolic increases and decreases at each time point but will be individually different in the degree of these changes. The property of a systematic within-subject change of pattern expression across conditions (beyond mere mean trends) is referred to as an “ordinal trend”. The number of subjects who violate the rule of increasing expression from baseline to follow-up can be used as a statistic to test the null-hypothesis of the absence of an ordinal trend in the data [7]. Monte-Carlo simulations of regional noise that is independently and identically distributed according to a Gaussian generate the p-level for the value of the number-of-exception criterion observed in our subject sample. The demonstration of a significant ordinal trend (p < 0.05) supports the claim that the pattern that was obtained in the longitudinal data reflected disease progression on a subject-by-subject basis, rather than a change on the mean across time points that came about as a result of overly influential subject outliers. Patterns resulting from multivariate analysis assign different weights to all voxels included in the analysis, depending on the salience of their covariance contribution. Voxel weights that are positive indicate a positive correlation between the subject expression value and the associated regional activity, whereas negative weights indicate a negative correlation. Thus, as the expression of a pattern increases, activity in the positively weighted regions increases as well, whereas activity in the negatively weighted regions decreases. The absolute magnitude of a regional weight determines the slope of this change: for instance, a region whose weight is twice as large as that of another also changes its activity twice as steeply. As in our earlier studies, the individual’s expression of the pattern at each session (time point) is quantified by the subject score. This value is obtained by the operation of an inner product (=covariance across brain regions) between the covariance pattern in question and a subject’s scan. It quantifies the extent to which a subject expresses the pattern in each scan as a single number, which can be used for further analysis. Change in pattern expression for each subject as a function of time is measured by the difference of that subject’s score from baseline to follow-up. The OrT analysis was conducted in the PMCI and SMCI group separately as well as in the combined PMCI and SMCI group. The change of the expression scores from the initial

C. Huang et al. / Neurobiology of Aging 28 (2007) 1062–1069

1065

Fig. 1. Selected ROIs in ROI-based analysis. Superior parietal lobe and posterior cingulated (left), medial temporal lobe (middle), parieto-temporal association cortex (right).

investigation to the second time point was correlated with the change of neuropsychological tests. Furthermore, we used a prospective voxel-based network quantification approach to quantify the expression of the derived pattern on a prospective group. These forward calculations were performed using a fully automated routine. 2.5.4. SPM analysis for longitudinal PMCI The longitudinal rCBF changes in PMCI subjects were estimated using the ‘paired t-test’ model. The baseline and follow-up scans were coded as two conditions. Contrasts were defined to examine both positive and negative longitudinal differences in blood flow. The SPM(T) maps were obtained at a height threshold of p = 0.001, uncorrected for the multiple comparisons. 2.5.5. VOI-based BRASS analysis for progression rates of PMCI Image registration and quantification were additionally performed in PMCI to calculate the annual progression rate using BRASS ROI-based approach developed by Nuclear Diagnostics Ltd., London, England [16]. The patient datasets were iteratively registered using nine parameter linear registration to a normal template using normalized mutual information as similarity function. The software uses a map of 46 VOIs. Eight VOIs, which are the most interesting volumes in early AD were selected to calculate the annual rCBF progression rate in PMCI. Those regions are bilateral superior parietal lobe, parieto-temporal association cortex, posterior cingulate and medial temporal lobe. The relative rCBF in PMCI was calculated as cerebellar ratios (mean value of region/mean value of bilateral cerebellar cortex). The selected ROIs are shown in Fig. 1. The annual progression rate in PMCI was calculated using the equation of (rCBFt1 − rCBFt2 ) × 100%/(rCBFt1 × Timefollowup ).

acquired from BRASS analyses. Post hoc analyses were performed with Scheff´e’s test. The gender differences among groups were evaluated using chi-square test. Logistic regression was applied and the receiver operating characteristic (ROC) curve was calculated for the evaluation of the diagnostic accuracy using SPECT and neuropsychological tests between PMCI and SMCI at baseline. The longitudinal changes of rCBF values acquired from BRASS and neuropsychological tests were evaluated separately in PMCI and SMCI using dependent t-test. Correlational analyses were performed by using Spearman’s correlation coefficient. The significance level of the above analyses were set up as p < 0.05.

3. Results 3.1. Neuropsychology 3.1.1. Group comparison of neuropsychology at baseline The results of group comparison of neuropsychological tests among PMCI, SMCI and controls at baseline are presented in Table 1. PMCI performed significantly worse concerning episodic memory, visuospatial function and MMSE as compared to SMCI at baseline. 3.1.2. Longitudinal changes of neuropsychological tests PMCI had significantly decreased MMSE and episodic memory during the follow-up (MMSE—baseline: 25.7 ± 2.6 (mean ± S.D.), follow-up: 23.2 ± 4.6 (mean ± S.D.), p = 0.0224; recognition words D —baseline: 2.1 ± 1.0 (mean ± S.D.), follow-up: 1.7 ± 0.9 (mean ± S.D.), p = 0.0137). No significant MMSE changes were detected in SMCI (MMSE—baseline: 27.4 ± 2.0 (mean ± S.D.), follow-up: 26.6 ± 2.6 (mean ± S.D.), p = 0.2031; recognition words D —baseline: 3.1 ± 1.2 (mean ± S.D.), follow-up: 3.2 ± 1.1 (mean ± S.D.), p = 0.5978).

2.6. General statistics

3.2. SPECT

Group differences were analyzed with ANOVA among PMCI, SMCI and controls at baseline concerning age, followup time, neuropsychological tests and relative rCBF data

3.2.1. SSM analysis at baseline PCA analyses were performed to derive the covariate patterns that differentiate PMCI versus SMCI (PS-pattern),

1066

C. Huang et al. / Neurobiology of Aging 28 (2007) 1062–1069

Table 1 Means, standard deviations and significances of neuropsychological tests of PMCI, SMCI and controls at baseline Cognition domain

Test

General cognitive function

MMSE

Episodic memory

Recognition words Recognition words D

Semantic memory

PMCI

SMCI

C

Significance [PMCI vs. SMCI vs. C]

25.7 (2.6)

27.4 (2.0)

29.4 (0.8)

3.5 (3.5) 2.1 (1.0)

0.7 (0.6) 3.1 (1.2)

0.5 (0.6) 3.7 (0.9)

Information Similarities

16.8 (5.4) 16.3 (8.0)

20.0 (4.8) 19.7 (5.0)

23.2 (3.8) 22.2 (3.4)

[PMCI vs. C] [PMCI vs. C]

Visuospatial

Block design

17.6 (6.0)

23.8 (8.0)

33.1 (5.7)

[PMCI vs. SMCI vs. C]

Attention

Digit symbol TMT-A-time TMT-B-time

31.0 (12.7) 61.6 (29.1) 179.4 (109.6)

35.0 (11.3) 56.0 (22.9) 146.5 (70.9)

48.1 (10.5) 36.0 (9.3) 84.2 (39.1)

[PMCI vs. C] [SMCI vs. C] [PMCI vs. C] [SMCI vs. C] [PMCI vs. C] [SMCI vs. C]

[PMCI vs. SMCI] [PMCI vs. C] [PMCI vs. SMCI] [PMCI vs. C]

Significance level: p < 0.05. PMCI: progressive mild cognitive impairment. SMCI: stable mild cognitive impairment. C: controls.

PMCI versus controls (PC-pattern) and SMCI versus controls (SC-pattern), separately. The results are shown in Fig. 2. PMCI displayed similar pattern as compared to SMCI and controls separately, which involved the decreased rCBF in parietal lobe and frontal lobe hyperperfusion in common. Cerebellum had increased rCBF in PMCI as compared to SMCI. The covariate pattern of SMCI as compared to controls showed a patchy pattern represented by the hypoperfusion in parietal lobe, parieto-temporal association cortex, cerebellum and frontal lobe, as well as increased rCBF in frontal lobe, insular cortex and subcortical regions.

The expression of PS-pattern had a weak negative correlation with visuospatial function (R2 = 0.12, p = 0.0433) in the combined group of PMCI and SMCI. The expression of PC-pattern had negative correlation with MMSE (R2 = 0.49, p < 0.0001), episodic memory (recognition words D ) (R2 = 0.25, p = 0.0022), semantic memory (Information test) (R2 = 0.22, p = 0.0043), visuospatial function (block design) (R2 = 0.41, p < 0.0001), attention (digit symbol) (R2 = 0.46, p < 0.0001), and positive correlation with attention tests: TMT-A-time (R2 = 0.44, p < 0.0001) and TMTB-time (R2 = 0.37, p = 0.0001) in the combined PMCI and

Fig. 2. The disease related covariate patterns using neuronal network analysis. Red/yellow represents the positive region weights (increased rCBF) in the pattern; blue/green represents the negative region weights (decreased rCBF) in the pattern. Left: PMCI related covariate pattern as compared with SMCI (PS-pattern), middle: PMCI related covariate pattern as compared with controls (PC-pattern), right: SMCI related covariate pattern as compared with controls (SC-pattern). PT, parieto-temporal association cortex, WM, white matter, C, controls. Color bar represents Z-score.

C. Huang et al. / Neurobiology of Aging 28 (2007) 1062–1069

1067

Table 2 Estimates and confidence interval of logistic regression and ROC analysis between PMCI and SMCI at baseline Model I SPECT PS-pattern Neuropsychology Recognition words test D Block design ROC (%)

−5.32 (−8.82, −1.83) – – 82

Model II – 1.01 (0.26, 1.75) 0.15 (0.02, 0.28) 84

Model III −8.64 (−14.66, −2.63) 1.78 (0.35, 3.21) – 92

ROC: represents the area under the receiver operating characteristic curve. PS-pattern: PMCI related covariate pattern as compared with SMCI at baseline.

controls. The SC-pattern expression scores showed relatively weak negative correlation with MMSE (R2 = 0.26, p = 0.0004), visuospatial function (block design) (R2 = 0.22, p = 0.0018), attention (digit symbol) (R2 = 0.17, p = 0.0066), and positive correlation with attention tests (TMT-A-time) (R2 = 0.17, p = 0.0083) in the combined SMCI subjects and controls. A logistic regression model was applied and ROC analysis was performed for the discrimination between PMCI and SMCI at baseline. The results are shown in Table 2. The expression scores of PS-pattern and neuropsychological tests had similar diagnostic accuracy of 82 and 84%, separately, between PMCI and SMCI at baseline. The combined rCBF measurement and neuropsychological tests increased diagnostic accuracy to 92%. 3.2.2. Ordinal trends (OrT) analysis for longitudinal data The results of OrT analysis in PMCI group are shown in Fig. 3. In a whole brain analysis of voxel activity, we found

the pattern displayed ordinal trend properties such that pattern expression increased from baseline to follow-up. MonteCarlo simulations of regional noise that is independently and identically distributed according to a Gaussian generate the p-level of p = 0.0225 for the value of the number-of-exception criterion observed in our subject sample. The covariance pattern was characterized by progressive decreases in the activity of hippocampus, parahippocampus, parietal lobe and brainstem, associated with increases in the medial frontal lobe, cingulate, and occipital lobe. No significant ordinal trend pattern was found in SMCI group. The expression of PMCI OrT pattern was prospectively calculated in SMCI group, which showed no significant expression of PMCI OrT pattern in SMCI group. 3.2.3. SPM analysis for longitudinal PMCI PMCI had decreased rCBF during the follow-up in parietal lobe (Talairach coordinates: [−38, −45, 61]), hippocampus ([30, −39, 2]), brainstem ([0, −35, −8]), temporal lobe ([−42, −26, −10]), and anterior cingulate ([2, 25, −8]), as

Fig. 3. (A) The topographic OrT pattern derived in PMCI longitudinal investigation. (B) The subject scores for the OrT pattern in PMCI, which showed increased OrT pattern expression over time in most of the subjects and continued elevation of network activity relative to baseline. Monte-Carlo stimulation test p-value less than 0.05. Color bar represents Z-scores.

1068

C. Huang et al. / Neurobiology of Aging 28 (2007) 1062–1069

Table 3 Means and standard deviations of relative rCBF values at baseline and follow-up, significances of the longitudinal rCBF differences and annual progression rate in PMCI ROI

Baseline

Follow-up

p

APR (%)

Left superior parietal lobe Right superior parietal lobe

0.84 (0.05) 0.84 (0.05)

0.83 (0.06) 0.82 (0.06)

0.0330* 0.0318*

1.4 (2.6) 1.7 (4.1)

Left parieto-temporal lobe Right parieto-temporal lobe

0.83 (0.05) 0.85 (0.05)

0.82 (0.06) 0.83 (0.06)

0.0518 0.0702

1.3 (2.9) 1.1 (2.8)

Left medial temporal lobe Right medial temporal lobe

0.84 (0.04) 0.84 (0.04)

0.81 (0.05) 0.81 (0.05)

0.0295* 0.0440*

2.6 (4.9) 2.9 (5.7)

Left posterior cingulate Right posterior cingulate

0.84 (0.07) 0.80 (0.08)

0.84 (0.09) 0.80 (0.09)

0.6917 0.9336

0.0 (10.5) 0.0 (9.4)

APR: annual progression rate. p: paired t-test between baseline value and follow-up. * p < 0.05.

well as decreased blood flow longitudinally in cerebellum ([−26, −58, −26]). 3.2.4. VOI-based BRASS analysis for progression rates of PMCI The baseline and follow-up mean values and standard deviations of PMCI in the selected ROIs are shown in Table 3. PMCI subjects had significantly decreased relative rCBF during the follow-up in bilateral superior parietal lobe and bilateral medial temporal lobe. The annual progression rates in these regions ranged from 1.4 to 2.9%.

4. Discussion The results of SSM analysis revealed the PMCI related covariate patterns, which differentiate the rCBF differences between PMCI and SMCI, as well as PMCI and controls at the initial investigation. The scores of the covariate pattern expression were associated with the decline of the cognitive function at baseline. The PMCI related pattern was represented by the hypoperfusion in parietal lobe associated with elevated rCBF in frontal lobe and cerebellum as compared with SMCI at baseline, which is consistent with our previous findings [10]. Neuropsychological deficits in the domains of episodic memory, visuospatial function and general cognitive function represented by MMSE were detected in PMCI group as compared with SMCI at the initial investigation. Mutivariate analysis of SPECT data and neuropsychological exams resulted in similar diagnostic accuracy between PMCI and SMCI at baseline, which are 82 and 84%, separately. Combining these two methods improved the diagnostic accuracy to 92%. Our findings at baseline are consistent with previous SPECT study using multivariate analysis [4]. Borroni et al. examined 31 MCI subjects with 2 year follow-up. The converters presented a pattern involving parietal and temporal lobe, precuneus, and posterior cingulate cortex. The combined memory function and rCBF identified 77.8% of the preclinical AD at baseline [4]. In the longitudinal study, the results of OrT analysis showed a covariate pattern in PMCI with existing ordinal

trend. The pattern involved the reduced rCBF in hippocampus, parahippocampus, parietal lobe and brainstem, which is consistent with SPM findings and previous report [13] as well as increased rCBF in frontal lobe, cingulate and occipital lobe. Most subjects showed elevated pattern expression during the follow-up (Fig. 3B). The three violations were tested by Monte-Carlo stimulation test and defined as random violations. Previous SPECT study in AD indicated that both the initial pattern of deficits and changes over time were heterogeneous, the probability of AD was 54% with bilateral temporal and/or parietal deficits, 69% with bilateral temporoparietal defects with additional deficits, 17% with no defects and 11% with frontal defects only [12]. This phenomena of the heterogeneity of AD might explain the derive OrT pattern was not found to be strong (p = 0.02) although significant. VOI-based approach detected significant rCBF decline in superior parietal lobe and medial temporal lobe in PMCI. The annual progression rate ranged from 1.4 to 2.9%. Parietal lobe showed to be a sensitive marker for the dementia occurrence and progression, in which regions, the rCBF of PMCI decreased at baseline as compared to SMCI and it continued to progress until the subjects reached the diagnostic point of dementia. It is necessary to point out that the MCI group was relatively young and that the generalizability to older MCI patients is not quite clear, which needs further studies in larger age-cohorts. Concerning the SMCI subjects, a patchy covariate pattern was derived in SMCI group as compared with normal subjects at baseline, which involved several cortical and subcortical areas. No OrT pattern was derived in SMCI group in the longitudinal study and the derived PMCI OrT covariate pattern did not express in SMCI group. The above findings are in agreement with the heterogeneous genesis of SMCI. Concerning the clinical outcome of SMCI, a subgroup of SMCI may progress to AD in the future, as evidenced by the findings in the SC-pattern at baseline that parietal lobe and parieto-temporal association cortex had decreased rCBF in SMCI group as compared with controls, but to a lesser degree as compared to PMCI. These

C. Huang et al. / Neurobiology of Aging 28 (2007) 1062–1069

regions are the typical areas involved in preclinical and early AD. Other SMCI subjects might be the preclinical dementia of other types. It is possible that some may never progress to any significant extent. Subjects with functional memory impairment induced by psychological disorders such as depression or anxiety disorder might improve after several months follow-up. Different pathogenesis related to different pathophysiological pathways of brain activity in SMCI group could result in multiple regions presented in the imaging study of SMCI, while analyzed as a single group. This could explain the patchy unstable SC-pattern derived by SSM analysis at cross-sectional study, as well as the fact that no OrT pattern was derived among SMCI subjects and SMCI as a group did not express PMCI OrT pattern. Moreover, increased rCBF in cerebellum was found in PMCI when compared with SMCI, but not when compared with controls. It indicates that the involved region might not necessarily relate to the progression of AD per se, but reflect the nature of SMCI as well as the heterogeneity and selection bias of the SMCI group. The results showed that the clinical heterogeneity of MCI is reflected in different patterns of rCBF and neuropsychological changes. Combined SPECT and neuropsychological testing might predict the future development of AD in patients with MCI. SPECT with multivariate technique and neuropsychological testing can be used objectively for the diagnosis of AD at baseline and to monitor the longitudinal changes of brain function in very early AD, for example, to evaluate treatment effects.

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12]

Acknowledgments This work was supported by the Swedish Medical Research Council, the Swedish Council for Social Research, the Gamla Tj¨anarinnor Foundation, the Greta LindenauHansell’s Foundation, Alzheimerfonden, the Karolinska Institute and the Swedish Society of Medicine, Nuclear Diagnostics Ltd, London, England and SADF. We are grateful to Toni Flanagan for the assistance in preparing the manuscript.

[13]

[14]

[15]

[16]

References [1] Almkvist O, Basun H, Backman L, Herlitz A, Lannfelt L, Small B, et al. Mild cognitive impairment—an early stage of Alzheimer’s disease? J Neural Transm Suppl 1998;54:21–9. [2] Backman L, Forsell Y. Episodic memory functioning in a community-based sample of old adults with major depression:

[17] [18]

1069

utilization of cognitive support. J Abnorm Psychol 1994;103(2): 361–70. Bennett DA, Wilson RS, Schneider JA, Evans DA, Beckett LA, Aggarwal NT, et al. Natural history of mild cognitive impairment in older persons. Neurology 2002;59(2):198–205. Borroni B, Anchisi D, Paghera B, Vicini B, Kerrouche N, Garibotto V, et al. Combined 99mTc-ECD SPECT and neuropsychological studies in MCI for the assessment of conversion to AD. Neurobiol Aging 2006;27(1):24–31. Folstein MF, Folstein SE, McHugh PR. Mini-mental state”. A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 1975;12(3):189–98. Habeck C, Hilton HJ, Zarahn E, Flynn J, Moeller J, Stern Y. Relation of cognitive reserve and task performance to expression of regional covariance networks in an event-related fMRI study of nonverbal memory. Neuroimage 2003;20:1723–33. Habeck C, Krakauer JW, Ghez C, Sackeim HA, Eidelberg D, Stern Y, et al. A new approach to spatial covariance modeling of functional brain imaging data: ordinal trend analysis. Neural Comput 2005;17(7):1602–45. Habeck C, Rakitin B, Moeller J, Scarmeas N, Zarahn E, Brown T, et al. An event-related fMRI study of the neurobehavioral impact of sleep deprivation on performance of a delayed-match-to-sample task. Cog Brain Res 2004;18:306–21. Holman BL, Johnson KA, Gerada B, Carvalho PA, Satlin A. The scintigraphic appearance of Alzheimer’s disease: a prospective study using technetium-99m-HMPAO SPECT. J Nucl Med 1992;33(2): 181–5. Huang C, Wahlund LO, Almkvist O, Elehu D, Svensson L, Jonsson T, et al. Voxel- and VOI-based analysis of SPECT CBF in relation to clinical and psychological heterogeneity of mild cognitive impairment. Neuroimage 2003;19(3):1137–44. Huang C, Wahlund LO, Svensson L, Winblad B, Julin P. Cingulate cortex hypoperfusion predicts Alzheimer’s disease in mild cognitive impairment. BMC Neurol 2002;2(1):9. Ishii K, Mori E, Kitagaki H, Sakamoto S, Yamaji S, Imamura T, et al. The clinical utility of visual evaluation of scintigraphic perfusion patterns for Alzheimer’s disease using I-123 IMP SPECT. Clin Nucl Med 1996;21(2):106–10. Kogure D, Matsuda H, Ohnishi T, Asada T, Uno M, Kunihiro T, et al. Longitudinal evaluation of early Alzheimer’s disease using brain perfusion SPECT. J Nucl Med 2000;41(7):1155–62. Moeller JR, Nakamura T, Mentis MJ, Dhawan V, Spetsieres P, Antonini A, et al. Reproducibility of regional metabolic covariance patterns: comparison of four populations. J Nucl Med 1999;40(8): 1264–9. Petersen RC, Doody R, Kurz A, Mohs RC, Morris JC, Rabins PV, et al. Current concepts in mild cognitive impairment. Arch Neurol 2001;58(12):1985–92. Radau PE, Slomka PJ, Julin P, Svensson L, Wahlund L-O. Automated segmentation and registration technique for HMPAO-SPECT imaging of Alzheimer’s patients. Medical imaging: Image processing. Proc SPIE 2000;3979:372–84. Wechsler D. Wechsler adult intelligence scale-revised manual. San Antonio: Psychological Corp.; 1981. American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, fourth ed. American Psychiatric Editors, Washington, 1994.