Longer-Term Investigation of the Value of ¹⁸F-FDG-PET and Magnetic Resonance Imaging for Predicting the Conversion of Mild Cognitive Impairment to Alzheimer’s Disease: A Multicenter Study

SEAD-J participants

The data set of SEAD-J was collated from nine institutions across Japan [6]. Patients with amnestic MCI were followed up annually for 5 years. A total of 114 patients (64 women and 50 men, mean age, 70.8±7.5 years) were enrolled in the study. This study was approved by the ethics committees of all participating institutions. Each patient signed an informed consent form after receiving full explanations of the procedures involved.

All patients were initially assessed with a series of neuropsychological tests, including the Mini-Mental State Examination (MMSE), the Alzheimer’s Disease Assessment Scale-cognitive component-Japanese version (ADAS-J cog), the Clinical Dementia Rating (CDR), the Geriatric Depression Scale (GDS), the Everyday Memory Check List (EMCL), and the Logical Memory Subset of the Wechsler Memory Scale Revised (WMS-R LM). In accordance with the inclusion criteria, patients with MCI were between 50 and 80 years of age, with an MMSE score ≥24, a GDS score ≤10, a WMS-R LM I score ≤13, a WMS-R LM II part A and part B score ≤8 (maximum = 50), and a CDR memory box score equal to 0.5. Patients with less than 6 years of formal education were excluded.

¹⁸F-FDG-PET

¹⁸F-FDG-PET scans were performed with the patient in a resting state, 40 to 60 min after venous injection of ¹⁸F-FDG (254±107 MBq). A static scan was performed for 10±5 min, either in 2D or 3D mode. Attenuation was corrected using either a transmission scan with segmentation for dedicated PET, and by a computed tomography (CT) scan for PET/CT.

The ¹⁸F-FDG-PET images were processed to produce 3-dimensional stereotactic surface projections (3D-SSP) and generate z-score maps, using iSSP (version 3.5, Nihon Medi-Physics, Tokyo, Japan). 3D-SSP was created with the Neurological Statistical Image Analysis Software (NEUROSTAT) developed by Minoshima et al. [7]. NEUROSTAT anatomically normalizes the individual PET data to the standard brain and compares the regional voxel data with the normal database, calculating the z-score ([normal mean _ individual value] / normal SD) for each voxel of the cerebral surface, and displays the sites at which voxel value is statistically reduced. The normal database was constructed from 50 control subjects (31 men and 19 women, mean age 57.6 years), with 10 subjects from each of the 5 participating institutions. All control subjects had no memory complaints and no history of neurologic or psychiatric disorders. The results of their neurologic examinations and brain imaging examinations (MRI or CT) were normal, and their cognitive function was judged to be normal by experienced neurologists (MMSE score,25 to 30).

Interpretation of PET images

Three experts, blinded to clinical information, independently assessed the reconstructed PET images. They correlated the 3D-SSP z-score maps to classify the images into the different dementia patterns according to the criteria of Silverman [8]. In these criteria, the findings of FDG-PET are classified into seven interpretive patterns as positive (P1–P3, P1+) or negative (N1–N3) for presence of a progressive neurodegenerative disease in general and AD specifically, compared with results of longitudinal or neuropathological analysis. In which P1 indicates progressive PET pattern consistent with presence of AD showing hypometabolism in parietal/temporal±frontal cortex, while P1+, P2, and P3 indicate progressive PET pattern but not consistent with AD (P1+, the presence abnormal findings other than P1; P2, frontal predominant hypometabolism; P3, hypometabolism of both caudate and lentiform nuclei). N1-N3 patterns indicate all negative scans, in which N1 represents normal metabolism. When the classification of the 3 raters did not match, the cases were discussed, and a consensus result was agreed upon. They referred to patients MRI images to exclude the partial volume effect due to severe atrophy and cerebral vascular lesion, and space-occupying lesions when a local hypometabolic area was observed with the PET image.

PET score

The AD t-sum was calculated, as described in previous publications [9, 10], by using the PALZ module in the PMOD software package (version 3.2, PMOD Technologies, Zurich, Switzerland). The AD t-sum indicates the severity of the metabolic decrease in those brain areas that are typically affected by AD (multimodal association cortices mostly located in the temporal and parietal lobes) and includes adjustment for age effects.

In this study, the AD t-sums were converted into a PET score dividing them by a previously determined upper normality limit and applying a log transformation to convert them into normally distributed values, according to the following formula [11]: PET score = log2 [(AD t-sum / 11,089) + 1].

MRI

All subjects were scanned with either a 1.5 T or 3T MRI system. T1-weighted 3-dimensional sagittal scans of the brain were acquired using a T1-weighted fast field echo sequence. Images were analyzed using the voxel-based specific regional analysis system for AD (VSRAD^® advance, Eisai, Tokyo, Japan), developed based on voxel-based morphometry (VBM) [12–14]. We used the latest version of VSRAD running VBM analysis on the basis of SPM8 plus diffeomorphic anatomical registration using exponentiated Lie algebra (DARTEL). This application automatically analyzed the 3D data as a series of anatomical standardizations, segmentations, and smoothings using these algorithms. Then each processed segmented image was compared to the mean and SD values of the gray matter or white matter images of healthy subjects using voxel-by-voxelz-score analysis. The database for the healthy controls of this software contained data from 40 men and 40 women aged between 54 and 86 years (mean age, 70.2±7.3). In this study, the average z-score in the target volume of interest (VOI) for medial temporal structures, including the entorhinal cortex, head to tail of the hippocampus, and amygdala, was used for further analyses.

Follow-up

Patients were assessed at 1-year intervals for5 years. The CDR, MMSE, EMCL, and WMS-R-LM tests were re-administered at each visit. Repeat ¹⁸F-FDG-PET and MRI scans were optional. Conversion to dementia was established when CDR became ≥1.0. No further follow-up of patients reaching CDR ≥1.0 was performed. AD was diagnosed when a patient fulfilled both CDR ≥1.0 and the National Institute of Neurological and Communicative Disorders–Alzheimer’s Disease and Related Disorders Association’s (NINCDS-ADRDA) “probable AD” criteria. The diagnosis of other diseases was based on established clinical criteria, including vascular dementia (VaD) [15], dementia with Lewy bodies (DLB) [16], and frontotemporal dementia(FTD) [17]. Two researchers of the SEAD-J group, blinded to the PET results, established the final clinical outcome of each patient based on the submitted case reports.

Logistic regression analyses

Multivariate logistic regression analyses were used to assess whether baseline ¹⁸F-FDG-PET, MRI, and/or neuropsychological tests were predictive of longitudinal clinical outcome. We estimated the odds of AD conversion versus nonconversion as a function of age, sex, educational level, WMS-R-LM I, WMS-R-LM II, MMSE, ADAS-J cog, GDS, PET score, PET visual interpretation, and VSRAD z-score (average, left, and right). Results were considered significant at p < 0.05. Statistical analyses were performed with SPSS (version 22.0; SPSS, Chicago, IL, USA).

Baseline characteristics

Of 114 patients, 42 dropped out from the study without a complete 5-years follow-up. Of the remaining 72 patients, 53 progressed to dementia, 49 developed AD, and four developed non-AD dementia (two patients with FTD, one with DLB, and one with VaD) (Fig. 1). Among the 49 AD converters, 41 (83.7%) converted within the first 3 years, and 4 patients (8.2%) converted in the fourth year and another 4 patients converted in the fifth year. The cumulative conversion rate over the 5 years was 72%.

Fig.1

Schematic summary of clinical outcomes in all patients with mild cognitive impairment (MCI). Originally, 114 patients with MCI were included. Forty-two patients dropped out during the 5 year. Our final sample size for the analyses was 68 patients (excluding 4 patients who converted to other types of dementia).

Table 1 shows the demographic and neuropsychological data from the initial visit of patients who progressed to AD (AD converters) and those who did not (nonconverters). At baseline, the two groups differed in WMS-R-LM, MMSE, ADAS-J cog, and GDS scores. There were no differences in age, educational level, or sex distribution between these two groups.

Interpretation of PET images

P1 and P1+ patterns were observed in 70.2% and 7.9%, respectively, of all amnestic patients with MCI at baseline. The remaining 21.9% patients presented other patterns, including the P2 in 4.4% cases. In this study, all P1+ patients showed a P1 pattern with occipital hypometabolism. Therefore, P1 and P1+ patterns were combined as an AD/ DLB pattern for the purpose of calculating diagnostic performance, as per a previous report [6].

The visual interpretation of PET images predicted conversion to AD during the 5-year follow-up with an overall diagnostic accuracy of 82%, a sensitivity of 94%, and a specificity of 53% for the 68 patients in this study. The diagnostic parameters for each follow-up interval are summarized in Table 2.

PET score

The mean PET score of the AD converters (1.16±0.67) was significantly higher than that of nonconverters (0.73±0.60) at baseline (p = 0.02). When mean PET scores were calculated according to conversion time, those who converted during the first and second years of the study had significantly higher PET scores than nonconverters. In contrast, those who converted in the third, fourth, and fifth years had no significant differences in mean PET score when compared with nonconverters (Fig. 2).

Fig.2

Box plot of baseline positron emission tomography (PET) scores according to conversion time. Patients with mild cognitive impairment progressing to Alzheimer’s disease in the first and second years had significantly higher scores than nonconverters (p < 0.05 and <0.01, respectively). 1st y, first year converter; 2nd y, second year converter; 3rd y, third year converter; 4th y, fourth year converter; 5th y, fifth year converter; Non, nonconverter.

The area under the curve of the receiver operating characteristic (ROC) analysis for the PET score was highest for the 2-year follow-up. The ROC-derived PET score thresholds obtained with the use of the Youden index [18] yielded an adjusted accuracy of 83%, with 70% sensitivity and 90% specificity at a threshold value of PET score = 1.03 at the second-year follow-up point. At the fifth-year observation, the accuracy was 60% with 53% sensitivity and 84% specificity. The diagnostic parameters for each follow-up interval are summarized in Table 2.

MRI

Within the MRI cohort, 6 patients were excluded from further analysis. Four patients at one particular institution were examined with a 3T MRI system, and 2 patients who converted did not undergo an initial MRI examination at their first visit.

For the AD converters and nonconverters, the bilateral mean z-scores in the target VOI for the medial temporal structures at baseline were 1.80±0.71 and 1.27±0.72, respectively (p = 0.01); the right side mean z-scores were 1.75±0.86 and 1.37±0.78, respectively (p = 0.10); and the left side mean z-scores were 1.80±0.86 and 1.03±0.75, respectively (p = 0.001). Therefore, the mean z-score in the target VOI of AD converters was significantly higher than that of nonconverters for the left medial temporal structures.

When the mean VSRAD z-scores were calculated according to conversion time, patients who converted in the first and second years had significantly higher z-scores than nonconverters (p = 0.004 and 0.016, respectively). In contrast, those who converted in the third, fourth, and fifth years had no differences in mean z-scores compared with nonconverters (Fig. 3).

Fig.3

Box plot of baseline voxel-based specific regional analysis system for Alzheimer’s disease (VSRAD) z-scores according to conversion time. Patients with mild cognitive impairment progressing to Alzheimer’s disease in the first and second years had significantly higher scores than nonconverters (p < 0.005 and <0.05, respectively). 1st y, first year converter; 2nd y, second year converter; 3rd y, third year converter; 4th y, fourth year converter; 5th y, fifth year converter; Non, nonconverter.

The area under the curve of the ROC analysis for the mean VSRAD z-scores was highest for the 5-year follow-up (Table 2). The ROC-derived thresholds using the Youden index for mean z-score yielded an adjusted accuracy of 73%, with 79% sensitivity and 63% specificity at a threshold value of z-score = 1.21 during the 5-year follow-up. The accuracy of VBM analysis presented little fluctuation through 5 years. When examining the left medial temporal structures only, the following results were obtained: accuracy; 76%, sensitivity; 77%, specificity; 68% (cut off; 1.14).

Neuropsychological tests

The diagnostic performance for neuropsychological batteries in each interval period are summarized in Table 2. Diagnosis sensitivity of WMS-R LM II and MMSE showed a high tendency, whereas the specificity of WMS-R LM I indicated a high tendency.

Logistic regression analysis

Multivariate logistic regression analysis identified educational level, WMS-R-LM II, ADAS-J cog, GDS, and PET visual interpretation as predictors distinguishing AD converters from nonconverters over 5 years (Table 3). When age, sex, educational level, WMS-R-LM I, WMS-R-LM II, MMSE, ADAS-J cog, GDS, PET score, PET visual interpretation, and VSRAD z-score were included in the stepwise procedure, AD conversion was significantly associated with educational level (p = 0.065; odds ratio, 1.34; 95% CI, 0.98–1.83), WMS-R-LM II (p = 0.003; odds ratio, 0.61; 95% CI, 0.44–0.85), ADAS-J cog (p = 0.104; odds ratio, 1.21; 95% CI, 0.96–1.52), GDS (p = 0.037; odds ratio, 1.65; 95% CI, 1.03–2.65), and PET visual interpretation (p = 0.026; odds ratio, 13.64; 95% CI, 1.36–136.52). The combination of PET visual interpretation, educational level, and these neuropsychological tests (WMS-R-LM II, ADAS-J cog, GDS) distinguished AD converters from non-converters with 87.9% accuracy, 89.8% sensitivity, and 82.4% specificity.

Conclusions

¹⁸F-FDG-PET visual assessment was shown to be highly effective in predicting conversion to AD from MCI. In particular, the combination of PET visual interpretation and neuropsychological tests could be a valuable biomarker for conversion detection in patients under longer-term observation. The semi quantitative PET score estimation method could be suitable for detection of rapid converters, but not for slow converters among MCI patients. In addition, structural MRI focusing on the medial temporal area showed a stable predictive value throughout the 5-year study interval.