Progress in neuroimaging of Alzheimer’s disease


Professor Hiroshi Matsuda, Department of Nuclear Medicine, Saitama Medical University Hospital, 38 Morohongo, Moroyama-machi, Iruma-gun, Saitama, Japan. Email:


The main purposes of neuroimaging in Alzheimer’s disease (AD) have progressed from diagnosis of advanced AD to diagnosis of very early AD at a prodromal stage of mild cognitive impairment (MCI), prediction of conversion from MCI to AD and differential diagnosis from other diseases causing dementia. Structural MRI studies and functional studies using FDG-PET and brain perfusion SPECT are widely used in diagnosis of AD. Outstanding progress in the diagnostic accuracy of these neuroimaging modalities has been obtained using statistical analysis on a voxel-by-voxel basis after spatial normalization of individual scans to a standardized brain-volume template instead of visual inspection or a conventional region of interest technique. In a very early stage of AD, this statistical approach revealed gray matter loss in medial temporal areas prominently in the entorhinal cortex and hypometabolism or hypoperfusion in the posterior cingulate cortex and precunei. These two findings might be related in view of anatomical knowledge that the regions are linked through the circuit of Papez. This statistical approach also offers a predictive value of conversion from MCI to AD and accurate evaluation of therapeutic effects on brain metabolism or perfusion. This development in functional and structural imaging might be an important surrogate marker for trials of disease-modifying agents.


With increasing life expectancy across the world, the number of elderly people at risk of developing dementia is growing rapidly, and Alzheimer’s disease (AD) remains the most common cause of dementia in all age groups. For almost three decades, positron emission tomography (PET) and single photon emission computed tomography (SPECT) have been used to investigate functional alterations of the brain in patients with AD. On the other hand, magnetic resonance imaging (MRI) has also been used to detect specific atrophy of the brain in AD. Recent advances in instruments have enabled us to investigate functional and morphological alterations in fine structures not only of cortical but also of subcortical areas with high spatial resolution. Moreover the development of computer assisted analysis using three-dimensional stereotactic surface projection (3D-SSP),1–3 easy Z-score imaging system (eZIS)4–6 and voxel-based morphometry (VBM)7,8 based on statistical parametric mapping (SPM)9 have afforded objective and more reliable assessments of functional and morphological abnormalities by means of stereotactic coordinates rather than visual interpretation of raw tomographic images. This stereotactic approach is a voxel by voxel analysis in the stereotactic space to avoid subjectivity and to adopt the principle of data-driven analysis. Although the alternative approach using a volume of interest (VOI) technique has gained general acceptance, it is limited by the fact that the selection of the sample depends on the observer’s a priori choice and hypothesis, and leaves large areas of the brain unexplored. Recent medications like cholinesterase inhibitors; for example donepezil, has been shown to be able to delay the progression of AD.10 This fact makes present studies on AD move toward earlier diagnosis and longitudinal investigations to assess therapeutic effects. Neuroimaging has been reported to be superior to cognitive testing for early diagnosis of AD.11 In this article I review recent progress in functional and morphological neuroimaging of AD using PET, SPECT and MRI.


We developed a software program running on Windows XP for automated diagnosis of brain perfusion SPECT and designated this program as Easy Z-score imaging system (eZIS; Fig. 1).4–6 In this program, voxel-based analysis was performed using a Z-score map calculated from a comparison of a patient’s data with the control database; Z-score = ((control mean) −(individual value))/(control S.D.) in the same manner as in a 3D-SSP method. Anatomical standardization of SPECT images into a stereotactic space was performed using SPM2. Therefore this program was made from the combination of 3D-SSP and SPM2. It has been reported that 3D-SSP with two-dimensional surface projection of cortical activities is less sensitive to artifacts derived from incomplete anatomical standardization of brain with localized cortical atrophy.12 However a 3D-SSP technique loses information on three-dimensional location that SPECT images inherently possess. This program also has an advantage in the capability to incorporate SPM results into an automated analysis of Z-score values as a VOI. A specific VOI can be determined by group comparison of SPECT images for patients with a neuropsychiatric disease with those for healthy volunteers using SPM.

Figure 1.

Easy Z-score imaging system (eZIS) at the very early stage of Alzheimer’s disease (AD). Easy Z-score imaging system analysis by comparison of a brain perfusion SPECT image for a 78-year-old man with probable AD (with MMSE of 24), with the mean and standard deviation SPECT images of healthy volunteers after normalization to global mean cerebral blood flow values. Gray scale Z-score maps ranging from 2.0 to 5.0 with extent threshold of 300 voxels were displayed by overlaying on transaxial sections and surface rendering of the spatially normalized MRI template. White lines (arrows) enclose a volume of interest (VOI) with the most significant decline of rCBF in very early AD obtained from group comparison with healthy volunteers by SPM2. High Z-score indicating significant reduction of regional cerebral blood flow was obtained within this VOI.

Even if a center can construct a normal database of good quality comprising a large number of healthy volunteers, other centers have not been able to use this normal database because of differences between the used gamma cameras, collimators and physical correction algorithms. Since SPECT exhibits greater variations in image quality among different centers than PET, conversion of SPECT images may be necessary for sharing a normal database. In our eZIS software, we incorporated a newly developed program making it possible to share a normal database in SPECT studies.4 A Hoffman 3-dimensional brain phantom experiment was conducted to determine systematic differences between SPECT scanners. SPECT images for the brain phantom were obtained using two different scanners. Dividing these two phantom images after anatomical standardization by SPM2 created a 3-dimensional conversion map. This conversion map was applied to convert an anatomically standardized SPECT image using one scanner to that using the other scanner. The SPM2 demonstrated that this conversion was of adequate validity in comparative analyzes of these SPECT images with different scanners.4 The present use of a conversion map obtained from SPECT images of the same phantom provided very similar SPECT data despite extreme differences between scanners. The present method may be useful for combining normal databases from different centers and could greatly enhance the diagnostic value of brain SPECT imaging by standardization of data analysis using a common normal database.


We also developed a software program running on Windows XP for the automated diagnosis of brain MRI and designated this program as a voxel-based specific regional analysis system for AD (VSRAD; Fig. 2).8 The acquired 3-dimensional T1-weighted sagittal MRI is reformatted to gapless transaxial images. Anatomical standardization fitted each individual brain to a standard template brain in 3-dimensional space using SPM2, so as to correct for differences in brain size and shape and facilitate intersubject averaging. In spatial normalization, only a 12-parameter affine transformation was used to avoid segmentation errors caused by partial-volume effects inherently created by warping. The normalized MRI is then segmented into gray matter, white matter, cerebrospinal fluid and other components using a modified version of the clustering algorithm, the maximum likelihood ‘mixture model’ algorithm using SPM2. The segmentation procedure involves calculating for each voxel a Bayesian probability of belonging to each tissue class based on a priori MRI information with a non-uniformity correction. The segmented gray matter images are then subjected to an affine and non-linear spatial normalization using a template of a priori gray matter. The spatially normalized gray matter images are smoothed with an isotropic Gaussian kernel 12 mm in full width at half maximum to use the partial volume effect to create a spectrum of gray matter intensities. The gray matter intensities are equivalent to the weighted average of gray matter voxels located in the volume fixed by the smoothing kernel. Regional intensities can therefore be taken as equivalent to gray matter concentration. Each gray matter image of the patients is compared to the mean and S.D. of gray matter images of healthy volunteers using voxel-by-voxel Z-score analysis after voxel normalization to global mean intensities. These Z-score maps were displayed by overlay on tomographic sections.

Figure 2.

Voxel-based specific regional analysis system for Alzheimer’s disease (VSRAD) at the very early stage of AD. VSRAD analysis by comparison of a gray matter image for a 64-year-old man with probable AD (with MMSE of 27), with the mean and standard deviation gray matter images of healthy volunteers after normalization to global mean intensity values. Gray scale Z-score maps ranging from 2.0 to 5.0 were displayed by overlaying tomographic sections of the spatially normalized MRI template. Gray lines (arrows) enclose a volume of interest (VOI) with the most significant decline of gray matter concentration in very early AD obtained from group comparison with healthy volunteers by SPM2. High Z-score of 3.42 indicating significant atrophy of parahippocampal areas was obtained within this VOI.


The context of subjects with memory complaints who do not yet match criteria for AD but who are at high risk of developing a full–blown dementia syndrome in the next few years has been recently noteworthy. This ‘at-risk’ state is commonly referred to as mild cognitive impairment (MCI).13 In this prodromal stage of AD, decreases in regional cerebral blood flow (rCBF) as well as glucose metabolism in the posterior cingulate gyrus and precuneus have been reported using PET or SPECT (Fig. 1).14–17 Statistical image analysis using 3D-SSP or SPM was used to make these observations. We could hardly distinguish a slight decrease of rCBF or metabolism in this area in patients with early AD by visual inspection, since metabolic activity or rCBF in the posterior cingulate gyrus is as high as primary visual cortex in normal individuals at rest.14 Reduced PET measures of glucose metabolism in AD remain even after accounting for partial volume effects; thus, it is more than just an artifact resulting from increased cerebral fluid space.18 We have previously reported superiority of 3D-SSP analysis over visual inspection in the discrimination of very early AD from controls using brain perfusion SPECT.19 In this report, Z-scores in the posterior cingulate gyrus and precuneus gave better discrimination accuracy of 86% between AD and healthy controls than that in medial temporal areas, parietal association cortices, or temporal association cortices.

The observation that metabolic reduction in this area predicts cognitive decline in presymptomatic persons indicates that the pathophysiologic process begins well before even mild or questionable dementia is recognized clinically.20,21 PET measures of glucose hypometabolism reflect decreased synaptic activity due either to loss or dysfunction of synapses,22 and regional metabolic deficits observed on PET may reflect projections from dysfunctional neurons in other brain lesions. In non-human primates, lesions of entorhinal cortex, which are the first to be affected in AD,23 cause significant and long-lasting metabolic decline in a small set of remote brain regions, especially in the inferior parietal, posterior temporal, posterior cingulate and associated occipital cortices, and posterior hippocampal regions.24 Minoshima et al. reported that epilepsy patients who have undergone medial temporal lobectomy have reductions in rCBF in the thalamus and posterior cingulate cortex.25 Mosconi et al. and Hirao et al. reported functional connectivity between the entorhinal cortex and posterior cingulate gyrus in AD patients using fluorodeoxyglucose PET and brain perfusion SPECT.26,27 These results suggest that rCBF or metabolic reduction in the posterior cingulate gyrus and precuneus indicates the earliest functional changes in AD as a remote effect. According to our longitudinal SPECT study,16 rCBF decrease in the posterior cingulate gyrus and precuneus became ambiguous as disease processed. This may be due to more stability of flow in this area than that of other cortical areas as part of the disease process.

The area of the posterior cingulate gyrus and precuneus is known to be important in memory.28 A PET study revealed activation of the retrosplenial area of the posterior cingulate cortex during the episodic memory encoding task.29 Clinical evidence of existence of brain tumor or arteriovenous malformation in the retrosplenial cingulate cortex supports the importance of this area in memory function.30,31 The retrosplenial cingulate cortex receives input from the subiculum and projects to the anterior thalamus, thus providing an alternative route between the hippocampus and thalamus. Medial temporal structures involved in memory receive anterior thalamic input directly via the cingulate bundle and indirectly through a relay in the retrosplenial cortex.31 This thalamocortical portion of Papez’ circuit may be important in memory,32 and lesions of the cingulum and retrosplenial cortex may cause memory dysfunction by disrupting this pathway.

The PET study also showed activation in the precuneus during the episodic memory retrieval task but not in the control or the semantic memory tasks.28 Little is known concerning either the functions or connectivity of the precuneus. Anatomical evidence indicates prefrontal, temporal, occipital and thalamic connections to the precuneus. Recent advances in functional MRI revealed significant contributions of precuneus to episodic memory retrieval.33

Neuropathological studies have provided detailed information about which specific brain regions are selectively affected in the earliest stage of AD. The initial neuronal lesions of the neurofibrillary tangles and neuritic plaques appear to occur in the entorhinal cortex, a portion of the anterior parahippocampal gyrus that receives projections from widespread limbic and associated areas and gives rise to the perforant pathway, the major cortical excitatory input to the hippocampus itself.23,34 Some layers of the entorhinal cortex undergo 40% to 60% neuronal depopulation even in the earliest phase of AD, when memory impairments and patient complaints are subtle and the symptoms do not reach the threshold for a diagnosis of AD.23 In our study,8 SPM2 revealed gray matter loss selectively in the bilateral entorhinal areas in very early AD in good agreement with these pathological changes in the amnestic stage of AD. Frisoni et al. reported the most significant gray matter loss in bilateral hippocampal/amygdalar complex using SPM99 in three very early AD patients with MMSE scores of over 24.35 Using SPM99, Chetelat et al. also reported significant gray matter loss predominantly affecting the hippocampal region and cingulate gyri in patients with amnestic MCI.36


Mild cognitive impairment comprises a heterogeneous group with a variety of clinical outcomes and its members are at risk for developing AD. The prediction of conversion from MCI to AD using the initial neuroimaging studies is an important research topic. A recent longitudinal fluorodeoxyglucose PET study reported a high predictive value of reduced uptake in the parietal association areas and a lower predictive value of that in the posterior cingulate gyrus.37 Mosconi et al. also reported that converters demonstrated reduced glucose metabolism in the inferior parietal cortex as compared with non-converters.38 These results strongly demonstrate the high predictive value of functional abnormality in the parietal association areas. However, two longitudinal studies suggested high predictive value of functional abnormality in the posterior cingulate gyrus.39,40

Our investigation based on a comparison of 52 converters with 24 non-converters from MCI to AD at 3-years of follow-up showed reductions of rCBF in the bilateral parietal areas and the precunei in converters as compared with non-converters.41 The logistic regression model revealed that reduced rCBF in the inferior parietal lobule, angular gyrus, and precunei has high predictive value and discriminative ability of converters and non-converters. Our data suggest that the initial rCBF SPECT studies of individuals with MCI may be useful in predicting who will convert to AD in the near future.

Stoub et al. reported determination of baseline entorhinal and hippocampal volumes and their rate of atrophy could predict the risk of incident AD.42 They used proportional odds models to assess the relationship between entorhinal and hippocampal size and risk of incident AD among 58 non-demented elderly people. All participants were followed with annual clinical evaluations and structural MRI scans for up to 5 years. Fourteen of 58 non-demented participants developed AD during the follow-up period. Both baseline entorhinal volume and its slope of decline were independent predictors of incident AD, but initial hippocampal size and its rate of decline were not, after controlling for entorhinal volume.

Killiany et al. also reported comparisons of MRI measures of the entorhinal cortex and the hippocampus for predicting who will develop AD.43 The volume of the entorhinal cortex differentiated the subjects from those destined to develop dementia with considerable accuracy of 84%, whereas the measure of the hippocampus did not. These findings are consistent with neuropathologic data showing substantial involvement of the entorhinal cortex in the preclinical phase of AD and suggest that, as the disease spreads, atrophic change develops within the hippocampus, which is measurable on MRI.