In general, brain function can be indexed by perfusion imaging. Arterial spin labeling (ASL) MRI measures brain perfusion entirely non-invasively (Roberts et al. 1994). This is accomplished by using the physical principles of MRI to magnetically label blood water as an endogenous tracer for blood flow. Furthermore, ASL-MRI can readily be performed in the same scan session as structural MRI. Previous ASL-MRI studies in FTD demonstrated frontal hypoperfusion, similar to findings using PET and SPECT (Du et al. 2006). Moreover, hypoperfusion in FTD remained significant even after accounting for brain atrophy and partial gray and white matter volume variations, implying that hypoperfusion can be independent of brain volume loss. However, the relationship between hypoperfusion and brain atrophy in FTD has not been investigated thoroughly.

All images were processed as described in detail previously (Johnson et al. 2005). In short, each PWI was co-registered to a reference EPI (without labeling), which in turn was co-registered to the corresponding T1-weighted image using affine transformations with 9 degrees of freedom and the normalized mutual information co-registration algorithm available in the Statistical Parametric Mapping (SPM2) program (Welcome Department of Imaging Neuroscience; London, UK). Furthermore, the T1-weighted images were segmented into probabilistic maps of gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF) using Expectation-Maximization Segmentation (EMS), also available as add-on in SPM2 (Van Leemput et al. 1999, 2001). Although we did not express CBF in absolute units, the differences between the two groups are still valid, especially since we accounted for global variations in CBF across the subjects.

Then the probabilistic tissue maps were blurred to the resolution of PWI to estimate the GM, WM and CSF content of each voxel in the corresponding PWI image. Partial volume effect (PVE) correction for CSF was calculated for each voxel on the perfusion image on the basis of the method by Müller-Gärtner et al. (1992), which assumes that all contributions to perfusion are from brain tissue and that CSF contributes nothing. Furthermore, to account for variable amounts of GM and WM in the voxels, it was assumed that perfusion of GM is 2.5 times the perfusion of WM (Roberts et al. 1994), leading to corrected PWI intensities according to: SI_corr=SI_uncorr / (GM+0.4*WM), where SI_corr and SI_uncorr are the corrected and uncorrected perfusion signal intensities, respectively, and GM and WM are the corresponding probabilities of gray or white matter volumes, respectively, with GM+WM+CSF=1. To constrain the analysis to GM perfusion (since the PWI signal of WM is usually very small and CSF contributes only noise), a threshold was applied that included only voxels with at least 75% gray matter and less than 25% other brain tissue, such as white matter and cerebrospinal fluid. Finally, to reduce high signal artifacts in PWI (i.e. signal from blood vessels), PWI signals above two standard deviations of the mean signal were capped to a maximum of two standard deviations above the SI_corr. For voxelwise tests the T1-weighted images were spatially normalized to a study-specific brain template. The study-specific template was created by averaging the spatially normalized T1-weighted images of all subjects in the study using an initial affine transformation with 12 degrees of freedom, followed by non-linear transformations with 10×9×10 discrete cosine basis functions. To test for regional brain atrophy, the T1-weighted images also were segmented, smoothed (6 mm isotropic Gaussian), and modulated according to the optimized VBM protocol as outlined by Good et al. (2001).

Following the t-tests, a concordant/discordant analysis was performed. For the concordant/discordant analysis of GM atrophy and hypoperfusion changes, we used a non-parametric multimodal correlation test as described in our previous report (Hayasaka et al. 2006). In brief, this test performs a statistic on the T-values of both GM atrophy, denoted here S_GM and T-values of hypoperfusion, denoted S_Perf. The analysis includes first, the design of combining functions of concordance or discordance between GM atrophy and hypoperfusion, followed by permutation tests to obtain the distribution of each combining function and to perform a statistic.

We defined the following combining functions:

Figure 1 shows a scatter plot of T-values of GM perfusion versus GM atrophy from a t-test comparing FTD to CN subjects. Also illustrated are the various combining functions at a statistical threshold of p=0.05 (FWE).

To determine the statistical distribution of voxels defined by each combining function, we performed permutation tests on the group of subjects and used the cluster mass of voxels as a test statistic. These tests were conducted using the Statistical non-Parametric Mapping (SnPM) package (http://www.sph.umich.edu/ni-stat/SnPM).

The level of statistical significance in imaging tests was set to p<0.05. We used the concept of false discovery rate (FDR) (Genovese et al. 2002) to account for multiple comparisons in separate imaging tests of brain atrophy and hypoperfusion. To account for multiple comparisons in concordance/discordance tests of brain atrophy and hypoperfusion, we used adjustments based on FWE, which are more stringent than those based on FDR.

Figure 2 shows the results from separate SPM group analyses of GM hypoperfusion (A) and GM atrophy (B) in FTD relative to CN subjects. Only regions that both T1-weighted and ASL perfusion MRI covered are shown. This indicates FTD patients had significant hypoperfusion bilaterally in the medial frontal cortex and right prefrontal cortex (p<0.05, FDR corrected). At the same statistical threshold level than perfusion, GM atrophy was by far more widespread in FTD, involving bilaterally large aspects of the medial frontal lobe and extending into the temporal lobes (p<0.05, FDR).

Figure 3 shows results from a concordance analysis of reduced GM perfusion and GM atrophy in FTD relative to controls. Concordance was found in the right prefrontal cortex and bilateral medial frontal lobe (p<0.05, FWE corrected).

We interpret the finding of discordant brain atrophy in absence of hypoperfusion in the premotor cortex as indication of brain tissue that is still functioning and connected in FTD, though other explanations are possible: First, loss of brain tissue other than neurons might contribute disproportionately to atrophy while surviving neurons still function normally as reflected by normal levels of perfusion. Indeed, some histopathological studies suggest that earliest cellular changes in FTD occur in astrocytes without frank neuron loss whereas neuronal damage is found in later stages of the disease (Broe et al. 2004; Kersaitis et al. 2004). This view is also supported by clinical observations that many patients with FTD do not have motor symptoms, despite apparent atrophy of the premotor cortex. Discordant MRI perfusion in presence of brain atrophy might provide an index for the condition of motor neurons in FTD which atrophy alone cannot provide. The recruited patients with FTD were all mild to moderate stage and accounting for disease severity, e.g. using MMSE, did not significantly alter the results. However, given the small number of subjects and limited power, we cannot exclude that the results might change for patients at a more advanced stage of FTLD. Further studies permitting concurrent assessment of structural and function alterations in FTD are necessary to test this conjecture.

Second, it is also possible that we lacked power with ASL-MRI to measure small alterations in hypoperfusion in presence of GM atrophy. A conservative power analysis of our data based on previous reliability tests (Jahng et al. 2005) suggests that the minimum difference in perfusion we can expect to detect with 28 patients and controls and 80% power (alpha=0.05) is about 20%. In comparison, the minimum difference in atrophy that we can detect at the same level of power and significance is 1.5%. This may also explain why we did not observe significant discordance between hypoperfusion without GM atrophy in FTD. In another study involving AD, however, we found regions with significant hypoperfusion in absence of atrophy, indicating that it is principally possible to measure discordant alterations between atrophy and perfusion with our method. Another issue that suggests technology rather than biological effects explains discrepancies between structural and perfusion findings are the typical observation of higher SNR of the perfusion signal from the frontal cortex, where regions are more adjacent to one another, compared to SNR of the perfusion signal from parietal brain regions. Hence, differences will be more datable in frontal than parietal brain regions. We also found concordant reductions of GM volume and perfusion in the right prefrontal cortex and bilateral medial frontal lobe, regions that have been consistently implicated in the pathology of FTD. Since the perfusion data were corrected for CSF and GM/WM tissue variations, the concurrent reductions in perfusion and GM volume cannot be explained simply as an artifact of partial volume effects. One possible explanation for concordant atrophy and hypoperfusion is that the remaining brain tissue is already damaged as indicated by reduced perfusion. It is interesting that the regions showing concurrent atrophy and hypoperfusion involved the right prefrontal cortex and bilaterally the medial frontal lobe. These regions are considered particularly vulnerable to FTD, as indicated in previous imaging studies (Broe et al. 2003). In addition, we found that concordance regions in the predominant right frontal lobe. It is consistent with several studies (Ishii et al. 1998; McMurtray et al. 2006; Du et al. 2006) that showed hypoperfusion and hypometabolism in the predominant right frontal lobe. The clinical significance of the metabolic and perfusion asymmetry is unknown. One possible explanation may be caused by sampling bias. Left-dominant patients can show language disturbances even in early stages. Thus, it is more likely that these patients are brought to neurology clinics. On the other hand, right-dominant patients may present predominantly with behavioral or psychiatric abnormalities, and so are first seen as psychiatric abnormalities, and potentially are misdiagnosed as having psychiatric illnesses. It should be noted that the equivalence between structural MRI and functional imaging, i.e. FDG-PET, in FTD has been seen in multiple studies of the two techniques independently but also in studies comparing the two modalities (Kanda et al. 2008; Kipps et al. 2009). Therefore, our findings of concordance, which agree with these previous studies, are not surprising. Nonetheless, the results support the equivalence between ASL measured hypoperfusion and PET measured hypometabolism in FTD.

The image analysis approach used in this report was previously employed to explore the concordance/discordance between structural and perfusion changes in patients with AD and mild cognitive impairment (Hayasaka et al. 2006). Our results demonstrate that the same concept applied to FTD provides new information, not available from testing each modality separately. The method could potentially be used to explore the relationship between various other imaging modalities, including PET and SPECT, in FTD and other neurodegenerative diseases.

Several limitations in our study ought to be mentioned. First, the patients were diagnosed clinically, without autopsy confirmation of FTD. Thus, it is possible that some of patients with FTD had other causes of dementia. Studies involving brain autopsies are needed to confirm that our findings are related to FTD pathology. The small number of patients limits power to detect significant differences in perfusion and generalization of the findings. Further studies including a larger number of patients are warranted to confirm the findings. A technical limitation is that this implementation of ASL-MRI in conjunction with the short lifetime of spin labels at 1.5 T did not permit covering more brain regions, especially the temporal lobes, which are also susceptible to FTD. Therefore, we may have missed other prominent regions with atrophy and hypoperfusion in FTD. Lack of perfusion data from the ventral prefrontal cortex and most of the temporal lobes limit, which both can be highly impacted in FTLD, limit the conclusions in this study. New volumetric ASL-MRI methods (Fernández-Seara et al. 2005; Günther et al. 2005) can provide more brain coverage, including the temporal lobe, especially when performed at a higher magnetic field strength than 1.5 T. Finally, another technical limitation is that the choice of combining functions is somewhat arbitrary and subjective and different choices for combining functions might alter the outcome. However, the sensitivity and robustness of the combining functions can be evaluated by simulations and permutation tests as we previously showed (Hayasaka and Nichols 2004). In addition, the functions we chose here are identical to those we used in previous studies of Alzheimer’s disease (Hayasaka et al. 2006) and brain aging (Zhang et al. 2008). Another limitation is that our concordance/discordance analysis is constraint to an analysis at the group level while examinations at the single subject level might be even more informative. In theory, a potential solution to this is performing the concordance/discordance analysis using each subject’s regional z-scores (deviation from the normal mean) rather than t-scores from the group comparison, supplemented by an appropriate statistic that takes both between group and across subject variations into account.

Soichiro Shimizu, Email: soichiro_s/at/hotmail.com.

Norbert Schuff, Email: Norbert.Schuff/at/va.gov.