European Journal of Nuclear Medicine and Molecular Imaging

, Volume 31, Issue 11, pp 1487–1494

Diagnosis of suspected Alzheimer’s disease is improved by automated analysis of regional cerebral blood flow

Authors

    • Department of Nuclear Medicine, Mont-Godinne University HospitalUCL-Université Catholique de Louvain
    • Department of Nuclear Medicine, Erasme HospitalFree University of Brussels
  • Satoshi Minoshima
    • Department of RadiologyUniversity of Washington
  • Jean George
    • Department of Nuclear Medicine, Mont-Godinne University HospitalUCL-Université Catholique de Louvain
  • Annie Robert
    • Departments of PHS Epidemiology and Biostatistics, Mont-Godinne University HospitalUCL-Université Catholique de Louvain
  • Christian Swine
    • Department of Geriatrics, Mont-Godinne University HospitalUCL-Université Catholique de Louvain
  • Patrice Laloux
    • Department of Neurology, Mont-Godinne University HospitalUCL-Université Catholique de Louvain
  • Thierry Vander Borght
    • Department of Nuclear Medicine, Mont-Godinne University HospitalUCL-Université Catholique de Louvain
Original Article

DOI: 10.1007/s00259-004-1597-7

Cite this article as:
Tang, B., Minoshima, S., George, J. et al. Eur J Nucl Med Mol Imaging (2004) 31: 1487. doi:10.1007/s00259-004-1597-7

Abstract

Purpose

Accurate diagnosis of Alzheimer’s disease (AD), the most common form of dementia, remains difficult. In order to assess whether fully automated stereotactic surface projection (3D-SSP) presentation contributes to the diagnosis of AD by single-photon emission computed tomography (SPECT), we investigated the diagnostic accuracy of transaxial display with and without 3D-SSP analysis as well as the correlation between cerebral perfusion in different cortical areas and the mini mental score (MMS).

Methods

Seventy-two patients referred because of cognitive impairment were included in the study. According to the National Institute of Neurological and Communicative Disorders and Stroke (NINCDS) and the Alzheimer’s disease and Related Disorders Association (ADRDA) criteria, 27 patients were diagnosed as having probable AD while 45 were classified as non-AD patients. 3D-SSP was used to quantify the regional cerebral blood flow (rCBF) acquired from SPECT imaging.

Results

Compared with the transaxial section presentation alone, 3D-SSP presentation improved the area under the receiver operating curve (p<0.05) as well as intra-observer (k=0.73 vs 0.88) and inter-observer (k=0.50 vs 0.84) reproducibility. Upon normalisation of regional to thalamic activity, multiple regression analysis revealed a strong correlation between the MMS and rCBF in the right parietal cortex (p=0.002).

Conclusion

Addition of 3D-SSP to the transaxial section display of ECD-SPECT studies improves the reproducibility and the diagnostic performance in respect of AD in patients with cognitive impairment and provides a valid tool for assessment of the severity of cortical perfusion abnormalities in such patients.

Keywords

Alzheimer’s diseaseSemi-quantificationCerebral blood flowMMSECD-SPECT

Introduction

Alzheimer’s disease (AD) is the most common cause of dementia (which accounts for 50–60% of cases of cognitive impairment) [1], and usually affects the elderly [2]. In 1990, the prevalence of AD was estimated to vary between 1% and 20% depending on the countries and the age of the population. Patients older than 80 years are more likely to be affected by the disease (prevalence 15–20%) compared with those aged 65–80 years (prevalence 2.5–6%) [3]. The constant “greying” of populations will have significant socio-economic effects because of age-related dementias. Yet, distinguishing AD from other causes of dementia remains a diagnostic challenge. Functional brain imaging is commonly used to guide the clinician.

Comparative studies between positron emission tomography (PET) and single-photon emission computed tomography (SPECT) [46] show that the diagnostic performance of the latter is slightly inferior, but still satisfactory for daily routine applications. The diagnostic accuracy reported in the literature ranges widely, depending on patient inclusion methodologies (i.e. comparison of AD patients with normal group of controls vs open trials) and the severity of dementia [7].

The three-dimensional stereotactic surface projection (3D-SSP) image analysis technique for SPECT is an automated voxel-by-voxel statistical analysis that evaluates semi-quantitatively the regional cerebral blood flow (rCBF) with ethylcysteine dimer (ECD) [8]. It was adapted from a method initially developed for PET analysis of the cerebral glucose metabolic rate.

In this study, we evaluated whether 3D-SSP combined with the standard transaxial display contributed positively to assessment of the AD cerebral perfusion pattern by SPECT among a cohort of patients referred to our clinic because of cognitive impairment. We also studied the relation between rCBF and the global level of cognitive function in these patients, as assessed by the mini-mental score (MMS).

Materials and methods

Study population

Seventy-two consecutive patients (47 females and 25 males) with cognitive impairment and a minimum follow-up of 2 years before assessment of the final diagnosis were selected. The mean (±SD) age of the population was 71±9 years (range 43–86) at the time of SPECT imaging. Patients referred to our clinic for evaluation of a slow cognitive decline underwent: (a) clinical assessment, (b) MMS psychometric testing, (c) blood tests, (d) a cerebral CT scan or MRI and (e) a SPECT study.

Twenty-seven of the 72 patients (18 women and nine men) were diagnosed as having probable AD according to the criteria of the National Institute of Neurological and Communicative Disorders and Stroke (NINCDS) and the Alzheimer’s Disease and Related Disorders Assocation (ADRDA) [9]. All AD patients were examined by a neurologist or a psychiatrist. The diagnosis of AD was made in the presence of gradually progressive cognitive deterioration without focal motor or sensory signs. Longitudinal assessment for 2 years confirmed the nature of the intellectual deficit. The 45 remaining patients suffered from non-AD-related cognitive impairment: mental disorders (n=22), Parkinson’s disease (n=6), probable frontotemporal dementia (n=4), cerebrovascular dementia (n=3), Steele-Richardson syndrome (n=1), chronic hydrocephalus (n=2), cerebral tumours (n=2), epilepsy (n=1), sleep apnoea (n=1), hypocalcaemic encephalopathy (n=1), Ménière’s disease (n=1) and CO intoxication (n=1). In the 27 probable AD patients, the MMS score was obtained in 24 cases within 1 month of the SPECT study (age 75±6 years, MMS 19±6).

SPECT imaging

Image acquisition

rCBF was measured with 99mTc-ethylcysteine dimer (99mTc-ECD, Neurolite, BMS, Princeton, USA). Acquisition was started 1 h after intravenous injection of 740 MBq of 99mTc-ECD under resting conditions using a triple-head camera (Trionix, Twinsburg, OH, USA) equipped with low-energy ultra-high resolution collimators. One hundred and twenty 15-s images in a 128×128 matrix were collected. Sinograms were reconstructed with a Shepp-Logan filter with a Hanning window. Neither attenuation nor scatter correction was used. The transaxial images (coronal, sagittal, axial) were displayed parallel to the canthomeatal line.

Image processing

The semi-quantitative analytical method of the rCBF is a fully automated method initially developed for PET by Minoshima et al. [10], which includes five main steps:
  1. Step 1

    Rotation and centering of the dataset. The mid-sagittal plane is determined iteratively by optimising the index of similarity between the right and left hemispheres.

     
  2. Step 2

    Realignment to the bicommissural line (AC–PC).

     
  3. Step 3

    Anatomical standardisation. The individual dimensions of the brain are deformed in order to adjust to the proportional grid system proposed by Talairach and Tournoux. The brain is stretched in three directions according to calculated linear scaling factors. After linear correction, the non-linear regional anatomical differences between the individual and the stereotactic atlas brain are minimised by a warping technique. From multiple stretching centre points, predominantly in the white matter, the stretching lines are drawn and are projected on the grey matter surface landmarks. For each stretching line, the algorithm measures a profile activity curve and matches the individual brain to the atlas brain.

     
  4. Step 4

    Data extraction using 3D-SSP. The cortical peak activity is projected onto the brain surface. Approximately 16,000 predefined surface pixels covering the entire surface of the brain are obtained. They are displayed on eight different aspects of the brain: one superior, one inferior, one anterior, one posterior, two lateral and two medial views.

     
  5. Step 5

    Statistical analysis of the normalised individual data. The data are normalised with respect to a reference structure, either the whole cortex (mean cortical activity) or the thalamus. The values of each patient are compared with a normal database originating from ten normal controls (mean age ± SD 64±7.5 years) obtained by courtesy of Bartenstein et al. [8]. A Z score is computed for each surface pixel using the following formula: Z score (x,y,z) = (Nmean(x,y,z)D(x,y,z))/Nsd(x,y,z) where Nmean(x,y,z) and Nsd(x,y,z) represent the mean and standard deviation of the normal database and D(x,y,z)), the individual value. The Z score (x,y,z) is also projected onto the pixel surface.

     

Image analysis

The typical scintigraphic pattern of AD was defined as bilateral reduced uptake in the cortical parieto-temporal areas sparing the neocortical primary areas, the basal ganglia, the thalamus and the cerebellum. This reduction in uptake could be asymmetrical and could extend to the frontal cortical area.

Two observers blinded to the final diagnosis were asked to grade the degree of certainty for the presence of an AD pattern employing a five-grade evaluation score: grade 1, definite AD; grade 2, probable AD; grade 3, intermediate; grade 4, probably no AD; grade 5, certainly no AD. The transaxial images were presented randomly to the observers, who were instructed to examine the images twice, with and without 3D-SSP analysis. In order to evaluate the intra- and inter-observer reproducibility, the same operation was repeated on two different days.

Data analysis

Considering that grades 1 and 2 correspond to a positive test for AD and that grades 3, 4 and 5 represent a negative test, we calculated the sensitivity, specificity and accuracy of each type of presentation. The kappa of Cohen was calculated to evaluate intra- and inter-observer reproducibility.

The diagnostic performance of the transaxial images alone or with the 3D-SSP analysis was evaluated using the receiver operating curve (ROC) technique (CORROC2 program) developed by Metz et al. [11]. This statistical program calculates the area under the ROC curve (Az), the standard deviation and the two-tailed p value for each observer and for each type of presentation (transaxial section and 3D-SSP). Data analysis was performed as follows: (1) Az comparison for each reader, transaxial section readings vs 3D-SSP (2) Az comparison for transaxial section readings between observer 1 and observer 2; and (3) Az comparison for 3D-SSP between observer 1 and observer 2. The level of significance was set at p<0.05.

In a predefined set of regions of interest, defined according to Brodmann areas in the stereotactic individual brain, 3D-SSP automatically measures the mean cerebral activity representing the rCBF. The rCBF value was normalised to the CBF of either the thalamus or the whole cortex. Associations between the normalised rCBF and the MMS score were examined by multiple regression analysis (SPSS program, Chicago, IL, USA, 1986).

Results

Impact of 3D-SSP on the clinical diagnosis of AD

For both observers, the ROC analysis demonstrated that 3D-SSP in association with the standard transaxial display yielded an improved diagnostic performance as compared with the standard transaxial display alone. A statistically significant increase in the area under the curve (Aztransaxial vs Az3D-SSP) was observed (Fig. 1a,b). No statistically significant differences between the diagnostic performances of the two readers were observed for the transaxial display (observer 1=0.77±0.06 vs observer 2=0.84±0.05) (Fig. 1c) or the 3D-SSP analysis (observer 1=0.91±0.05 vs observer 2=0.92±0.04) (Fig. 1d).
Fig. 1a–d

Improvement of the clinical diagnosis of Alzheimer disease by 3D SSP. a, b ROC analysis for the two observers (a observer 1, b observer 2; open circles, transaxial display; closed circles, 3D-SSP analysis) demonstrated an improvement in the diagnostic performance with 3D-SSP in comparison to the interpretation based on standard transaxial sections alone. c, d No significant differences were found between observers 1 and 2 for the transaxial (c) or the 3D-SSP analyses (d) (open circles, observer 1; closed circles, observer 2)

Sensitivity, specificity and accuracy were all improved by 3D-SSP (Table 1). Similarly, inter- and intra-observer reproducibility was improved, as shown by the values of the kappa coefficient for concordance (Table 2). Five patients in whom the state of dementia was unclear, with doubtful scores on psychometric testing at the time of the scan (mean MMS 26, range 25–28), later fulfilled the criteria for AD.
Table 1

Performance of ECD-SPECT for the diagnosis of ADa

 

Reader 1

Reader 2

TI

3D-SSPI

TII

3D-SSPII

TI

3D-SSPI

TII

3D-SSPII

Sensitivity (%)

59

81

48

78

52

70

59

70

Specificity (%)

89

93

82

91

93

98

91

98

PPV (%)

76

88

62

84

82

95

80

95

NPV (%)

78

89

73

87

76

85

79

85

Accuracy (%)

78

89

69

86

78

88

79

88

aSensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and accuracy were evaluated by two independent observers (nos. 1 and 2) on transaxial sections alone (T) or with 3D-SSP at two different time points (I and II)

Table 2

Intra- and inter-observer reproducibility

SPECT Images

Intra-observer (reader 1)

Inter-observer (reader 1 vs reader 2)

Axial sections

0.73

0.50

3D-SSP

0.88

0.84

Values are the kappa coefficient for concordance

Figures 2 and 3 illustrate the ECD-SPECT findings in two AD patients: the former has a typical AD pattern on both transaxial and 3D-SSP displays, whereas the latter has an AD pattern only clearly visualised on 3D-SSP presentation.
Fig. 2a–c

Typical AD pattern on both transaxial and 3D-SSP displays. The typical bilateral parieto-temporal deficit on both transaxial (a) and 3D-SSP presentations (b and c): right lateral (R. LAT), left lateral (L. LAT), superior (SUP), inferior (INF), anterior (ANT) and posterior (POST) views are shown. b 3D-SSP images of the rCBF projected onto the brain surface; c negative Z score display as the map of the reduction of the individual data compared with a normal control group

Fig. 3a–c

AD pattern only visualised on 3D-SSP presentation. AD patient in whom typical perfusion abnormalities unsuspected on transaxial display (a) were revealed on 3D-SSP analysis (b and c): right lateral (R. LAT), left lateral (L. LAT), superior (SUP), inferior (INF), anterior (ANT) and posterior (POST) views are shown. b 3D-SSP images of the rCBF projected onto the brain surface; c negative Z score display as the map of the reduction of the individual data compared with a normal control group

Correlation between rCBF of the parietal right brain normalised to the CBF of the thalamus and the MMS

Besides the evaluation of the usefulness of 3D-SSP techniques for correct identification of the AD pattern, we addressed the presence of a correlation between the cognitive level assessed by MMS and the semi-quantitative data for the cerebral blood flow measured by the program. Upon normalisation of regional to thalamic activity, multiple regression analysis revealed the strongest correlation between the MMS and rCBF in the right parietal cortex (r=0.61; p=0.002) (Fig. 4, Table 3). Similar correlations were found, in descending order, in the left parietal, right temporal, right occipital and left temporal cortices (Table 3). As the rCBF in each of the latter cortical regions correlated significantly with that in the right parietal cortex, only the correlation between the MMS and right parietal perfusion could be considered statistically valid (Simpson paradox). Regional activity normalised to the mean cortical activity provided a less significant correlation between MMS and the cortical rCBF (Table 4).
Fig. 4

Regression between rCBF in the right parietal cortex normalised to that in the thalamus and the MMS score. Using the activity normalised to the thalamus, multiple regression analysis showed the strongest correlation between the MMS and rCBF in the right parietal cortex (n=24, r=0.609, p=0.002)

Table 3

Multiple regression analysis between MMS and rCBF normalised to the thalamic activitya

MMS

R. Par.

L. Par.

R. Temp.

L. Temp.

R. Occ.

L. Occ.

R. Front.

L. Front.

r-Pearson

0.609**

0.531**

0.509*

0.406*

0.416*

0.381

0.252

0.246

p-value

0.002

0.008

0.011

0.049

0.043

0.066

0.235

0.247

Par. parietal, Temp. temporal, Occ. occipital, Front. frontal

Table 4

Multiple regression analysis between MMS and rCBF normalised to the mean cortical activitya

MMS

R. Par.

L. Par.

R. Temp.

L. Temp.

R. Occ.

L. Occ.

R. Front.

L. Front.

r-Pearson

0.496*

0.519**

0.441*

0.251

−0.016

−0.045

0.216

0.155

p-value

0.014

0.009

0.031

0.237

0.941

0.834

0.310

0.468

Par. parietal, Temp. temporal, Occ. occipital, Front. frontal

Discussion

The incidence of AD is growing with the aging of the population. In this study, we report on the contribution of 3D-SSP to ECD-SPECT as a valid and reproducible tool for assessing the diagnosis of AD. The interpretation of ECD-SPECT with 3D-SSP analysis in addition to the transaxial display is shown to be superior to interpretation based on the transaxial display alone. Additionally, the strong correlation between the right parietal CBF and the MMS score in AD patients suggests the potential of this technique for follow-up studies.

The “gold standard” for the in vivo diagnosis of AD has yet to be defined. Clinical history, psychometric testing, CT scan and functional imaging are all routinely used in patients with cognitive impairment. However, the NINCDS–ADRDA criteria, which are essentially evaluated in academic institutions, provide an accuracy rate as variable as 66–90% [1215]. The differences between these clinical trials are most probably due to the study design (open vs non-open trials) and the subjective evaluation by the clinician. Including a functional imaging assessment in the patient’s routine examination for dementia will serve to support and/or confirm the clinical observations.

Since the non-cortical brain regions, namely the cerebellum [16, 17] and the thalamus [1822], are known to be relatively preserved in AD, normalisation of regional cortical activity expressed in rCBF to that in extracortical areas seems rational. In this context, we have shown that rCBF of the right parietal region normalised to thalamic activity is highly correlated with the MMS score, better than when using global normalisation. Similarly, using a computer-assisted mapping program for SPECT to determine CBF in pathological aging, Karbe et al. [23] showed that regional normalisation to cerebellar activity provided a better indication of cortical impairment than did normalisation to global cerebral perfusion. The choice of the area for normalisation of regional cortical uptake is crucial for proper assessment of patients with suspected AD. As an alternative to 3D-SSP, the statistical parametric program (SPM 99, The Wellcome Department of Neurology, London, UK) is a software that integrates image analysis with a global normalisation procedure (i.e. scaling to average whole brain activity) [24]. Soonawala et al. [25] clearly showed that cerebellar normalisation, using additional software to SPM, renders the latter more accurate for the identification of patients with AD. 3D-SSP has the advantage of automatically normalising regional to thalamic activity without supplementary software manipulations. Additionally, 3D-SSP appears more suitable than SPM for standardisation of atrophied brains [26]. The correlation found between the uptake in the parietal cortex and the MMS score is consistent with most reports in the literature [2730]. Region-specific associations have also been found with 3D-SSP between the left temporo-parietal rCBF reduction and language performance, and between the right temporo-parietal rCBF reduction and praxis [31], while Jagust et al. found a significant relation between perfusion in the right parietal area and survival [32]. It may be concluded that the use of voxel-based evaluation analysis with both 3D-SSP and SPM could efficiently contribute to the early detection of AD [33].

In this study, we have considered as typical of AD the most commonly adopted scintigraphic pattern of AD, namely bilateral reduction in uptake in the cortical parieto-temporal areas, with sparing of the neocortical primary areas, sub-cortical structures and cerebellum [3437]. However, in mildly demented patients the AD pattern may be limited to one hemisphere. The use of conservative scintigraphic criteria for AD may have decreased our sensitivity. In addition, visualisation of the AD pattern on transaxial images remains difficult, especially when abnormalities are subtle. In early stages of disease or in cases of mixed pathologies, as are frequently observed among older people [3840], the difficulty in deciding for or against the presence of an AD pattern becomes more pronounced. rCBF semi-quantification applied to SPECT appears to improve the accuracy of AD diagnosis. The three-dimensionality of the 3D-SSP images of AD patterns facilitates correct diagnosis as compared with interpretation based on the standard transaxial section. Recently published data from Hanyu et al. [41] confirm our results. The authors of this Japanese article report that 3D-SSP displays better sensitivity and specificity than transaxial imaging in the diagnosis of AD. Our findings extend their results using ROC analysis, but also demonstrate a correlation between the rCBF and the level of global functioning as assessed by MMS. Hence, in comparison with transaxial image presentation, 3D-SSP significantly improves inter- and intra-observer reproducibility as well as the detection of rCBF abnormalities typical of AD in patients referred because of cognitive impairment.

Hypometabolic anomalies in the parieto-temporal regions have been reported in APOE-4 homozygote patients with a family history of AD and suffering from mild cognitive disorders [4244]. However, none of these studies clearly identified that these patients developed AD in the long run. Minoshima et al. have shown later development of AD in a group of eight patients who initially had a mild memory disorder associated with a decrease in metabolic activity in the cingulate cortex on 3D-SSP PET [45]. In our series, five patients with mild cognitive impairment at the time of the scan (mean MMS 26, range 25–28) later fulfilled the criteria for AD. Four had repeated ECD-SPECT 24±7 months later (range 17–34 months). Decreased rCBF in the association cortices was demonstrated in three, while rCBF was unchanged in one. The latter patient had the shortest time interval between scans. These preliminary data suggest that 3D-SSP ECD-SPECT may provide early diagnosis of AD in patients suffering from mild cognitive impairment.

The sensitivity of FDG-PET analysed with 3D-SSP for AD diagnosis has been found to be slightly superior to our SPECT results using ECD. Techniques, population and functional imaging criteria may account for such differences. Only a few studies have directly compared glucose metabolism and blood flow as measured with PET and SPECT techniques in healthy volunteers and AD patients. Ishii et al. [6] compared ECD-SPECT with FDG-PET in the same ten AD patients. They showed reduced perfusion in the parieto-temporal area in eight of the ten patients and reduced metabolism in nine. Using 99mTc-hexamethylpropylene amine oxime (HMPAO) for SPECT and SPM for the voxel-based analysis, Herholz et al. reported that the distinction between healthy volunteers and AD patients is more robust with PET than with SPECT [46]. However voxel-based analyses of FDG-PET and ECD-SPECT in the same AD patients are still lacking, and ECD-SPECT appears slightly superior to HMPAO-SPECT for the diagnosis of AD [47]. Nevertheless, considering the high cost of PET and its limited accessibility in daily clinical practice, SPECT remains a valuable tool that is easily combined with clinical evaluations for routine applications in older patients with cognitive impairment.

In conclusion, 3D-SSP improves the reproducibility and the diagnostic performance of ECD-SPECT for detection of the AD pattern in patients with cognitive impairment. This fully automated technique also provides a valid tool for assessing the severity of cortical perfusion abnormalities in patients with AD.

Acknowledgements

This study was supported by the Department of Nuclear Medicine, Mont-Godinne University Hospital.

Copyright information

© Springer-Verlag 2004