Herein we describe an exploratory study aimed at obtaining concurrent ASL-MRI and FDG-PET images using a novel protocol in a small group of patients with AD and age- matched controls, as these modalities have never been directly compared. The concurrent imaging protocol was well tolerated. The most striking finding was the similarity upon visual inspection of the patterns of CBF on ASL-MRI and metabolism on FDG-PET, as depicted in . The finding that perfusion and metabolism images resemble one another so closely suggests that these processes remain at least grossly coupled in the brain of Alzheimer’s patients. We have characterized this similarity further through qualitative expert analysis and simple whole-brain quantification. Qualitative analysis yielded similar sensitivity and specificity for AD detection for both ASL-MRI and FDG-PET in this small sample, though there was only moderate to good interobserver and intermodality agreement. Quantitative measurement of CBF and FDG SUVs generated ROC curves with AUCs of approximately 90% for both modalities. The 2 standard deviation confidence interval surrounding the difference between the two methods (calculated using the method of (23
)) included only small advantages in either direction -- at most a 10.4% accuracy advantage for FDG or a 13.8% advantage for ASL. Thus, this pilot study provides initial data suggesting considerable qualitative and quantitative similarity between ASL-MRI and FDG-PET in AD.
The qualitative similarity of the regional distribution of decreased CBF and diminished FDG uptake in images obtained concurrently in our study suggests that CBF closely parallels glucose metabolism in AD brain. However, the intermodality agreement (kappa statistic) was in the moderate range for both readers, suggesting that while there is considerable similarity between the two modalities, they can produce some diagnostic disparities. This is emphasized by the fact that significant frontal lobe hypoperfusion was noted on FDG-PET, but not on ASL-MRI, a finding which conflicts with a previous report (14
). The fact that both readers had extensive experience with FDG-PET but not with ASL-MRI images may explain why the intraobserver reliability and diagnostic confidence was lower for ASL-MRI, but the quantitative analysis was nearly identical. Readers may apply pattern recognition derived from years of FDG-PET analysis to ASL-MRI images, which are not identical. Further research and clinical experience will be needed to more formally define the patterns of change on ASL-MRI that associate with AD, and to determine the etiology of the heterogeneity that exists between ASL-MRI and FDG-PET.
Discrepancies between ASL-MRI and FDG-PET images could have two potential sources: image artifact, or true perfusion-metabolism mismatch. Perfusion imaging modalities such as ASL-MRI are susceptible to artifact in the setting of vascular compromise, such as arterial stenosis. We found no difference between groups in white matter ischemic changes and focal infarcts on the FLAIR MRI images, but this method is not perfect, and fixed perfusional deficits which did not lead to FLAIR-positive lesions might explain some of these discrepant diagnoses. The readers reported difficulty interpreting several ASL-MRI images due to small areas of hypoperfusion in the parieto-occipital junction, which could be confused with biparietal hypometabolism (). This may be due to decreased perfusion in the watershed zone between anterior and posterior arterial circulations. They also felt that it was more difficult to thoroughly examine the inferior temporal lobes, possibly due to diminished image quality from temporal bone artifact. Despite these issues, regional scoring of temporal lobe image intensity was similar between ASL and FDG, suggesting that these artifacts had minimal impact on diagnosis. It appears likely that these and perhaps other limitations of ASL-MRI images will need to be considered by expert readers, and that accurate diagnosis of AD in larger populations will require diagnostic criteria that are specific to ASL-MRI, and not simply adopted from FDG-PET. The qualitative similarity between ASL-MRI and FDG-PET images suggest that perfusion and metabolism are matched to some degree in AD, but no quantitative measures of regional perfusion or metabolism were made in this study to examine this definitively. Such efforts are in progress. A mismatch between oxygen uptake and glucose metabolism has been recently described in AD-susceptible brain regions in both normal controls and AD patients, and has been attributed to increased aerobic glycolysis in these vulnerable areas (25
). It remains to be seen if regional discrepancies between ASL-MRI and FDG-PET are due to this phenomenon.
Figure 4 Parieto-occipital hypoperfusion artifact on ASL-MRI. ASL-MRI images (left) from patient 9 (control) show areas of hypoperfusion in the parietal-occipital junction (arrows) that were not present on FDG-PET images (right). These were present in 4 control (more ...)
For the purposes of this report, we chose to focus on image analysis methods that might be accessible to the real-world radiologist, such as visual interpretations and basic whole-brain CBF quantification, as the primary advantages of ASL-MRI over FDG-PET center around convenience, cost, and clinical accessibility. Only a few groups have examined global CBG as a predictor of AD. Asllani et al. showed a 40% decrease in global grey matter-corrected CBF in AD patients (10
), while Yoshiura et al showed a 24% decrease without segmentation (13
). Our study showed a 31.8% decrease in whole brain CBF (without grey matter correction) in AD, suggesting that our simple, clinically-applicable method is sufficiently sensitive to capture global hypoperfusion in AD. Future work should also compare regional CBF and FDG uptake at the voxel level.
This study has several limitations. It was an exploratory study intended to evaluate a novel imaging method for the concurrent acquisition of FDG-PET and ASL-MRI data, and to determine if FDG-PET and ASL-MRI produce comparable images when performed concurrently in AD patients. The study was not prospectively powered to provide definitive sensitivity and specificity data for these modalities in AD, nor to determine statistical superiority of one modality (or equivalence, although we note that the confidence interval surrounding the AUC difference is fairly narrow). Extrapolating from our sample, if we wanted to narrow this confidence interval down to a worst case of a 5% advantage for FDG, a future study would need to increase our sample size by a factor of 3.11 (40 controls and 56 AD patients). ASL-MRI imaging of approximately 250 patients including patients with AD, MCI, and controls will occur as part of the second phase of the Alzheimer’s Disease Neuroimaging Initiative (ADNI). These data will allow more rigorous statistical appraisal of the diagnostic efficacy of this modality (27
). The current work provides further motivation for such rigorous comparison of FDG PET with ASL MRI.
The diagnosis of AD, used as the gold standard in this study, was made on clinical grounds, and can thus not be definitively confirmed without autopsy data, though ASL-MRI data in autopsy confirmed AD has recently been reported (28
). Conversely, some recent studies suggest that changes in cerebral metabolism in cognitively normal patients may precede the clinical appearance of cognitive deficits (29
), therefore some control subjects could also have had presymptomatic AD. Finally, the sample contained only AD patients and controls, and the results thus cannot be directly applied to the clinical setting, in which a mix of neurodegenerative and dementing illnesses are present.
In summary, we have compared FDG-PET and ASL-MRI images in a population of AD patients and cognitively normal controls using a concurrent protocol to ensure optimal head to head comparison. Our findings suggest that in this small but well-defined population, ASL-MRI yields very similar image patterns and comparable diagnostic accuracy as FDG-PET. While analysis of ASL-MRI images was slightly less consistent than that of FDG-PET, quantitative analysis for both modalities was nearly identical. Considering the many potential advantages of ASL-MRI, our results further bolster the potential utility of this image modality for the diagnosis of AD, and reinforce the need for further efforts to standardize and validate ASL-MRI methods in larger populations of mixed neurodegenerative conditions.