Regional correlation between resting state FDG PET and pCASL perfusion MRI

Abstract

To determine how arterial spin labeling (ASL) measured perfusion relates to baseline metabolism, we compared resting state cerebral perfusion using pseudo-continuous ASL and cerebral glucose metabolism using ¹⁸F-FDG PET in 20 normal volunteers. Greater regional metabolism relative to perfusion was observed in the putamen, orbitofrontal and temporal lobes, whereas perfusion was relatively higher in the hippocampus and insula. In a region of interest analysis limited to gray matter, the overall mean correlation between perfusion and metabolism across voxels was r=0.43 with considerable regional variability. Cross-voxel correlations between relative perfusion and metabolism in mean ASL and PET images of all 20 subjects were the highest in the striatum (caudate: r=0.78; putamen: r=0.81), and the lowest in medial temporal structures (amygdala: r=0.087; hippocampus: r=−0.26). Correlations between mean relative perfusion and metabolism across 20 subjects were the highest in the striatum (caudate: r=0.76; putamen: r=0.58), temporal lobe (r=0.59), and frontal lobe (r=0.52), but very poor in all other structures (r<0.3), particularly in caudal structures such as the hippocampus (r=−0.0026), amygdala (r=0.18), and insula (r=0.14). Although there was good overall correlation between perfusion and glucose metabolism, regional variability should be considered when using either ASL or ¹⁸F-FDG PET as surrogate markers for neural activity.

Introduction

Both regional cerebral perfusion and metabolism are closely coupled to neural activity and are important surrogate markers of brain function. Methods used to measure cerebral blood flow (CBF) include ¹⁵O-labeled water positron emission tomography (¹⁵O-PET), dynamic contrast enhanced magnetic resonance imaging (MRI), and arterial spin labeling (ASL). ¹⁸F-fluorodeoxyglucose (FDG) PET remains the standard method for measuring cerebral glucose metabolism (CMR_Glc). Radioactively measured perfusion and metabolism by ¹⁵O-PET and ¹⁸F-FDG PET, respectively, show overall good correlations in health and disease and similarly decline with age.^{1, 2, 3} Functional activation studies comparing CBF using ¹⁵O-PET and CMR_Glc using ¹⁸F-FDG PET have also shown excellent correlations between these two modalities.⁴

There has been growing interest, however, in the use of MRI-based methods using nonradioactive tracers to measure CBF, particularly with ASL. ASL perfusion MRI uses magnetically labeled protons in arterial blood water as a natural tracer to measure cerebral perfusion^{5, 6}. In ASL, arterial blood is labeled at the base of the brain by a radiofrequency-induced change in the longitudinal magnetization of protons in blood water. These magnetically tagged protons are then allowed to flow into the brain during the post-labeling delay time for the measurement of perfusion without the use of radiation. There is good correlation between ASL and ¹⁵O-PET in both resting state and activation studies, but a further advantage of ASL is the capability of obtaining a high-resolution structural image during the same scanning session of functional image acquisition, allowing for better co-registration.^{7, 8}

A growing body of evidence suggests that CMR_Glc measured by ¹⁸F-FDG PET and CBF measured by ASL also have reasonable correlations, but there are potentially different sensitivities to vascular artifacts and higher relative signal intensities in some brain regions with one method versus the other. Newberg and colleagues performed concurrent ASL measurements and ¹⁸F-FDG uptake during a visual stimulation task to show the correlation between mean CBF (mCBF) and mean CMR_Glc (mCMR_Glc). The correlation was high within the middle ranges of perfusion and metabolism but dropped off at higher perfusion rates.⁹ ASL was more sensitive in showing changes in the visual cortex driven by the stimuli, but PET showed relatively higher signal in the basal ganglia. Some of these differences may be explained by the difference in the time scale of the two modalities, as ASL measurements may fluctuate over a span of minutes, whereas PET uptake time is steady over 30–40 minutes followed by a static measurement.

Resting state ASL has also been used in direct comparison with ¹⁸F-FDG PET in a variety of neurological disorders ranging from dementia, epilepsy, brain tumors, and HIV, indicating growing clinical interest in applications of ASL.^{10, 11, 12, 13} However, there are no studies to date that have investigated the baseline correlation between resting state ASL measured CBF and resting state ¹⁸F-FDG PET measured CMR_Glc in non-diseased brain. This information may be important in determining whether the differences in perfusion or metabolism seen in either activated or diseased states should be attributed to the variable of interest or to baseline differences. In particular, a direct comparison may also reveal technical limitations of each modality. A map of the regional differences in the correlation between resting state CMR_Glc and CBF would be of utility in interpreting future studies that use ASL as a marker of neural function.

To address the relationship between resting CMR_Glc and CBF, we performed resting state pseudo-continuous ASL (pCASL) and ¹⁸F-FDG PET in 20 normal subjects and measured the correlation across different brain regions as well as across subjects. Pseudo-continuous ASL has the advantage of being technically more feasible than continuous ASL (CASL) while providing a higher signal to noise ratio than pulsed ASL (PASL).^{14, 15} It merges the advantages of the high labeling efficiency of PASL with the higher signal-to-noise ratio of CASL and has been shown to exhibit good reproducibility in both normal adults and children.^{16, 17, 18}

Materials and methods

Subjects provided written informed consent for protocols, which were approved by the Institutional Review Board and the Medical Radiation Safety Committee of the University. Experimental protocols were explained to the subjects and were performed according to the Declaration of Helsinki guidelines. Participants were chosen based on negative screening for neurological disorders and major medical illnesses. All subjects were normotensive with systolic blood pressure <140 mm Hg and diastolic blood pressure <90 mm Hg, and a resting heart rate between 60 and 100 beats per minute. PET and fMRI imaging were performed on different days in all but three subjects. Twenty participants (five male) with a mean age of 42.4 years (range 23–59 years) finished all components of the study.

Image acquisition

PET scans

Subjects were required to fast for >3 hours before undergoing the PET scan. Subjects with a postprandial finger stick glucose measurement of >140 mg/dL were automatically excluded. After injection of 5mCi (185MBq) of ¹⁸F-FDG into an antecubital vein, subjects sat with their eyes closed in a quiet dark room for a 40-minute uptake period. They were then scanned in 3D acquisition mode with a 30-minute emission scan (six blocks of five minute acquisitions) followed by a 10–20 minutes transmission scan on a Siemens ECAT EXACT HR+ Scanner. Sixty-three slices in the axial plane were obtained with a resolution of 1.6 mm × 1.6 mm × 2.4 mm. Images were corrected for scatter, decay, scanner dead time, and attenuation and were reconstructed by back projection using an all pass ramp filter of 2 mm full width half-maximum (FWHM). Attenuation correction was performed using a ⁶⁸Germanium source.

Pseudo-continuous arterial spin labeling (pCASL)

Subjects were scanned on a 3 Tesla Siemens Magnetom Trio scanner with their eyes closed. They wore video goggles while being scanned, which also tracked eye and lid movements. Twenty-four 5-mm-thick slices with a 1 mm gap were acquired in the transverse orientation with 40 pairs of labeled and unlabeled (control) measurements (80 measurements total) for each slice in an inferior to superior direction. An RF pulse train of 1,500 ms was applied 9 cm beneath the center of the acquired slices, with a mean gradient of 0.6 mT/m. A gradient-echo echo planar imaging (EPI) sequence was used for image readout. Other specifications were as follows: post-labeling delay=1.2 seconds, FoV=220 mm, matrix=64 × 64, voxel size=3.44 × 3.44 × 6 mm³, flip angle=90 degrees, rate-2 GRAPPA, TR=4,000 ms, TE=11 ms for a total scan time of 5 minutes 32 seconds.

Structural images

For image alignment, all subjects underwent a magnetization prepared rapid acquisition gradient echo (MPRAGE) structural scan with specifications as follows: 192 slices at 1 mm slice thickness, voxel size: 1.0 mm × 1.0 mm × 1.0 mm, FoV: 256 mm, flip angle: 9 degrees, TR=1,900 ms, TE=3.25 ms for a total scan time of 7 minutes 7 seconds.

Image processing

Image preprocessing was performed with Statistical Parametric Mapping (SPM) 8 (http://www.fil.ion.ucl.ac.uk/spm) and Matlab 2010 (Mathworks, Natick, MA, USA). Mean cerebral blood flow images for each subject were calculated using a perfusion reconstruction program developed in-house. Raw EPI images were realigned to the first image using SPM8 followed by pair-wise subtractions of labeled versus control images. The mean difference images were converted into CBF images following a standard perfusion model assuming a blood T1 of 1,650 ms and a labeling efficiency of 0.85.¹⁰ pCASL CBF images for each subject were co-registered with the corresponding structural MRI by maximizing their mutual information, normalized to the standard T1 MNI152 template, and smoothed in space using a 2 mm FWHM kernel.

Cerebral metabolic rate of glucose (CMR_Glc) images, which combined data from the six acquisition blocks of the PET scan, were made for each subject. These mean PET images were normalized to the MNI PET template in SPM8 using affine transformation and nonlinear warps. A brain mask was generated by skull-stripping normalized MPRAGE images using the BET program in each subject. The brain mask was then applied on the normalized pCASL CBF and PET CMR_Glc images. A gray matter mask thresholded for probabilities above 90% was applied for both CBF and CMR_Glc images before subsequent analyses. Both CBF and CMR_Glc maps were scaled to a global mean activity of ‘1' by dividing voxel intensities by the mean intensity of the gray matter, thereby generating relative CBF (rCBF) and relative CMR_Glc (rCMR_Glc) images, respectively. A final image of 68 slices with a voxel size of 2 mm × 2 mm × 2 mm was used for calculating correlation coefficients between rCBF and rCMR_Glc.

Statistical analysis

Paired t-tests were performed to compare rCBF and rCMR_Glc signal intensities for the gray matter-masked brain in all 20 subjects. The resulting t-map was thresholded at a whole brain false discovery rate (FDRp) corrected P<0.05 with a minimum cluster size of 30 voxels. The final map was overlaid onto a standard T1 template.

To specify regional correlations, the normalized rCBF and rCMR_Glc images were compared within 12 representative region of interests (ROIs) based on the automated anatomical labeling (AAL) toolbox (the frontal, parietal, occipital, and temporal lobes, cingulate cortex, thalamus, caudate, putamen, insula, amygdala, hippocampus, and cerebellum).

For cross-voxel correlations, mean rCBF and mean rCMR_Glc maps were generated by averaging the respective images from the 20 subjects. The pair of mean rCBF and mean rCMR_Glc maps were compared on a voxel-by-voxel basis within each ROI using Pearson's correlation coefficients. Since the number of voxels in each ROI was different, we performed Fisher's z transformation of r-values using the formula: Fisher's z=arctanh(r), which was further converted into the standard z score by dividing Fisher's z score with the standard error (1/√(ke−3)). For cross-subject comparison, the mean rCBF and rCMR_Glc values were extracted from the 12 ROIs in each subject, which were correlated across the 20 subjects using Pearson's correlation coefficients. Further, the mean rCBF and rCMR_Glc values of the 12 ROIs were compared using paired t-tests.

Histograms of rCBF and rCMR_Glc signal intensities were calculated from the mean maps of each modality. To minimize the potential effects of susceptibility artifacts on ASL CBF in lower slices, only the upper 40 slices were used for the histogram analysis.

Results

Subjects

Twenty subjects underwent PET imaging between 3 and 6 hours after their last meal. Fasting glucose measurements ranged from 61 to 109 mg/dL with an average of 85 mg/dL.

Mean intensity comparison

Normalized rCMR_Glc (PET) and rCBF (ASL) maps of the entire brain were averaged across subjects to show the overall common and different spatial patterns between the two modalities (Figure 1). PET and ASL images were highly consistent with each other after scaling the gray matter mean intensity to ‘1' in each modality. By visual appearance, ASL delineated cortical structures more clearly than PET, suggesting greater spatial resolution with ASL.

Figure 1.

Mean rCBF (ASL) and mean rCMRGlc (PET) from 20 normal subjects. CBF images are normalized to a standard MNI152 T1 template and CMR_Glc images to a MNI PET template and are presented in sequential slices from the vertex to the skull base.

A calculation of the signal-to-noise ratio based on the ratio between the mean gray matter signal and the standard deviation of white matter signal was 24.6 for CBF and 27.9 for PET.

A t-statistic map of the differences between rCMR_Glc and rCBF illustrates areas that show relative differences in PET and ASL signal intensities, with red values indicating areas of relatively greater PET signal and blue values indicating areas of relatively greater ASL signal (Figure 2). The orbitofrontal lobes, temporal lobes, and left putamen showed relatively greater metabolism compared with perfusion. In contrast, the bilateral hippocampus and insula showed greater ASL signal (Table 1).

Figure 2.

Comparison between rCMR_Glc and rCBF after gray matter masking. The voxel-wise paired t-test, thresholded at FDR corrected P<0.05 and cluster size>30 voxels. Positive values are shown in red, representing areas of higher relative CMR_Glc signal. Negative values are shown in blue, representing areas of higher CBF.

Table 1. Paired t-test comparisons of mean rCMR_Glc and rCBF.

	x	y	z	ke	T	Localization
rCMRGlc > rCBF
Cluster 1	12	20	−26	130	10.56	Right superior orbital gyrus
Cluster 2	−8	16	−26	106	10.49	Left rectal gyrus
Cluster 3	−46	−22	−30	85	10.48	Left inferior temporal gyrus
Cluster 4	−16	16	−6	90	9.83	Left putamen
	−12	10	−10		9.73	Left putamen
	−22	10	−8		9.64	Left putamen
Cluster 5	−54	−40	−28	45	9.61	Left inferior temporal gyrus
Cluster 6	52	−12	−40	46	9.58	Right inferior temporal gyrus
rCBF > rCMRGlc
Cluster 1	28	−14	−16	280	15.72	Right hippocampus
	32	−32	−2		13.15	Right hippocampus
	34	−24	−10		11.39	Right hippocampus
Cluster 2	48	−8	32	59	14.51	Right postcentral gyrus
Cluster 3	−28	−20	−14	230	13.58	Left hippocampus
	−26	−36	4		11.78	Left hippocampus
	−34	−26	−12		9.76	Left hippocampus
Cluster 4	−34	−22	16	57	11.28	Left insular lobe
Cluster 5	36	−10	18	154	10.1	Right insular lobe
	36	−22	16		9.94	Right rolandic operculum

Clusters of significant differences between gray matter rCMR_Glc and rCBF using a voxel-wise paired t-test, thresholded at FDRp corrected P<0.05 and cluster>30 voxels.

Cross-voxel correlation

Voxel-by-voxel correlations between mean rCMR_Glc and mean rCBF of the 20 subjects were calculated for the 12 ROIs. The scatter plots of ASL to PET signals in each ROI are presented in Figure 3A, showing an overall mean correlation coefficient of r=0.43. The highest regional correlations were seen for the striatum (caudate r=0.78 and putamen r=0.81) with the lowest correlations seen in medial temporal regions, such as amygdala (r=0.087), hippocampus (r=−0.26), and occipital lobes (r=0.12) (Table 2). As the voxel number in each ROI was different, we performed Fisher's z transform of r values and the results were not affected (Table 2).

Figure 3.

Regional comparisons between rCMR_Glc and rCBF. ROI-specific correlation plots made on a voxel-by-voxel basis between the mean rCMR_Glc and rCBF images averaged from all 20 subjects (A), and mean rCMR_Glc and rCBF correlations across the 20 subjects (B). Data reflect values from gray matter signal only. Colors indicate regions only. Blue: Upper brain regions; Red: middle brain regions; Green: Lower brain regions.

Table 2. Voxel-by-voxel correlation of mean rCBF and rCMR_Glc across 12 AAL ROI regions.

Region	ke	Correlation coefficient	Fisher's z	Standard z	P-value (two-tailed)
Frontal	21 018	0.41	0.44	64.00	0
Parietal	4239	0.50	0.55	35.75	0
Occipital	6400	0.12	0.12	9.77	0
Temporal	18 584	0.67	0.81	110.32	0
Cingulum	4902	0.61	0.70	49.16	0
Thalamus	762	0.45	0.49	13.44	0
Caudate	1285	0.78	1.04	37.27	0
Putamen	963	0.81	1.13	34.94	0
Insula	2862	0.63	0.74	39.48	0
Amygdala	371	0.087	0.087	1.67	0.095
Hippocampus	1565	−0.26	−0.27	−10.54	0
Cerebellum	14289	0.33	0.34	40.69	0

ke represents number of voxels in each ROI. Fisher's z transform was performed on the correlation coefficients to adjust for ke differences.

Cross-subject comparison

The correlation between mean rCBF and rCMR_Glc values across 20 subjects in the 12 ROIs was calculated to determine intersubject variability. The overall correlation was intermediate (mean r=0.19), with the highest correlations in the caudate (r=0.76), putamen (r=0.58), temporal (r=0.59), and frontal lobes (r=0.52). Correlations were the lowest in the hippocampus (r=−0.0026), parietal lobes (r=−0.083), occipital lobes (r=−0.05), cerebellum (r=−0.07), and amygdala (r=−0.18) (Figure 3B and Table 3). Table 3 (column 3 and 4) presents paired t-test comparisons of mean rCBF and rCMRGlc values of 12 ROIs in the 20 subjects.

Table 3. Cross subject correlation (column 1 and 2) and paired t-test (column 3 and 4) of rCBF and rCMR_Glc in 12 AAL ROI regions.

Region	correlation coefficient	P-value (two-tailed)	t-test two-tailed t-value	t-test two tailed p-value
Frontal	0.52	0.020	−2.56	0.02
Parietal	−0.083	0.73	1.88	0.08
Occipital	−0.22	0.36	3.60	0.00
Temporal	0.59	0.0058	1.79	0.09
Cingulum	0.29	0.22	−4.61	0.00
Thalamus	0.20	0.39	−2.32	0.03
Caudate	0.76	0.00	−0.77	0.45
Putamen	0.58	0.0074	7.67	0.00
Insula	0.14	0.56	−8.42	0.00
Amygdala	−0.18	0.45	−0.72	0.48
Hippocampus	−0.0026	0.99	−11.32	0.00
Cerebellum	−0.28	0.23	1.90	0.07

Histogram of PET and ASL

To determine whether rCBF and rCMR_Glc signals fell within a normal distribution, histograms of rCMR_Glc and rCBF were made for the 12 ROIs using slices 28–68. The pooled histograms are shown in Figure 4. Although the overall scaled values equaled ‘1,' the distribution of rCMR_Glc and rCBF for these 12 areas showed a normal distribution for rCBF but a skewed distribution for rCMR_Glc.

Figure 4.

Distribution of rCBF and rCMR_Glc values of gray matter in 12 AAL ROIs. Pooled histogram of (A) rCBF and (B) rCMR_Glc values from 12 predetermined ROIs, masked for gray matter signal.

Discussion

ASL has become an increasingly popular technique because of the ease of acquisition and its noninvasive nature. Given the differences in the methodology and biophysical mechanisms between ASL and PET, there is unlikely to be a complete correlation between these modalities. Understanding regional variations in correlations between brain perfusion and metabolism is important, as both perfusion and metabolism measurements are increasingly being used as complementary methods for measuring brain function. The correlation between CBF and CMR_Glc could be affected by regional differences in reserve oxygen extraction capacity, the basal arterial flow rate, the health of the brain tissue, and whether one method is more/less susceptible to artifacts in different brain areas. Any measurement of the correlation between perfusion and metabolism that is measured during a task or in a diseased brain state may not reflect the baseline correlation between the two methods.

The data in this study show that though there is an overall good correlation between CBF measured by ASL and CMR_Glc measured by ¹⁸F-FDG PET, there is considerable regional variability. Correlations between mean rCBF and mean rCMR_Glc across voxels were the highest in the caudate and putamen and were the lowest in medial temporal structures like the amygdala and hippocampus. Evidence for these regional differences indicates that a global comparison of perfusion to metabolism is unlikely to be useful, as neurological diseases usually show high regional vulnerability.

Previous ASL perfusion studies on subjects with mild cognitive impairment (MCI) and early Alzheimer's disease (AD) have shown hyper-perfusion in medial temporal regions, whereas FDG PET and structural MRI studies have reported hypo-metabolism and atrophy of medial temporal structures.^{19, 20, 21, 22} Although the mechanism underlying the seemingly ‘uncoupled' perfusion and metabolism in medial temporal structures of normal subjects is not clear, our findings are consistent with past ASL and PET studies in MCI and AD but shows that some of this discrepancy may exist as a normal phenomena, though accentuated in disease.

When comparing rCBF and rCMR_Glc across 20 subjects, a similar pattern of regional variation emerged as in the cross-voxel analysis. Cross-subject correlation was the highest in the striatum (caudate and putamen) and the lowest in the amygdala, hippocampus, parietal lobe, thalamus, cerebellum, and insula. The remaining brain regions revealed intermediate correlations between CBF and CMR_Glc. We found that the rCMR_Glc was significantly higher than rCBF in the putamen, which was also reported by Newberg et al in 2005 (reported more generally as the basal ganglia)⁹. These observations suggest that perfusion and glucose metabolism vary co-linearly in the striatum with metabolism being significantly higher than perfusion.

The observed regional differences between CBF and CMR_Glc should be interpreted within the context of specific physiological processes of each modality. ASL measures signal from water molecules that transit through the arterioles, capillaries, and into the brain tissue. The arterial transit time for the labeled blood to flow from the tagging region to the brain tissue is on the order of 1–2 seconds in healthy subjects.²³ The post-labeling delay of 1.2 seconds favors capturing signal from the arteriole side. The majority of ASL signal should be from extracellular space.²⁴ In contrast, PET captures signal emanating intracellularly, as FDG is taken up by cells by the glucose transporter, but is trapped once it becomes phosphorylated.²⁵ The temporal scales for tracer uptake and scanning procedures are drastically different for ASL and FDG PET, with the former on the order of seconds to minutes while the latter on the order of 30 minutes to an hour. All above factors may contribute to the observed regional differences between CBF and CMR_Glc.

Some of these regional differences between rCBF and rCMR_Glc may also be attributed to the technical limitations of each scanning method. The reduced rCBF relative to rCMR_Glc observed in the orbitofrontal and temporal regions is likely due to the susceptibility effects in gradient-echo EPI acquisitions, despite using a relatively short echo time (TE=11 ms). In contrast, greater rCBF relative to rCMR_Glc in the hippocampus and insula may arise from arterial transit effects or residual labeled signal in arteries in ASL scans. Recently, fast 3D sequences such as GRASE and SPIRAL have been employed for acquiring ASL signals.^{15, 26} Relying on an RF pulse train of spin-echoes, these 3D sequences are less sensitive to susceptibility artifacts than 2D gradient-echo EPI. It will be important to compare these 3D pCASL CBF scans with ¹⁵O-PET or ¹⁸F-FDG PET in future studies.

Finally, we observed that the histogram of rCMR_Glc was skewed relative to that of rCBF, despite all individual data being scaled to its own global signal intensity. For ASL, labeled spins can be modeled as a binomial process B(n,p), and the relaxation of bulk magnetization can be modeled as a Poisson process P(λ=np). In this case, n is on the order of Avogadro's number, whereas p is on the order of blood T1 (in PET, n is the tracer concentration and p is on the order of tracer half life). When dealing with PET data, each voxel can be treated as an independent Poisson random variable.²⁷ Even after corrections for scatter, decay and attenuation, a Poisson or a shifted Poisson model is still valid.²⁸ λ is an order of magnitude greater in ASL than PET, leading to an approximately normal distribution in ASL and a Poisson distribution in PET. This agrees with the results of Asllani et al.,²⁹ though they used CASL.

To minimize problems with partial volume averaging, we only used data from voxels that exceeded a high gray matter threshold. We minimized smoothing to 2 mm of the ASL data, which was obligatory given the 2 mm zoom on the PET acquisition. Nevertheless, partial volume effects may still be relevant in our analysis, especially in areas of tissue interface close to the subarachnoid space.

A limitation to our study is that CBF and CMR_Glc were not measured concurrently leaving open the possibility that some brain areas may show higher day-to-day variability in either perfusion or metabolism than others. However, despite this limitation, the voxel-by-voxel correlations for brain regions other than the medial temporal area were all above 0.40 and were particularly robust for the striatum. Mitigating against this weakness is the high scan-rescan reliability of ASL similar to other resting functional modalities and variability that is averaged out over multiple subjects.^{16, 17, 30} A design such as in Newberg's study where the FDG was injected while the subject was in the MRI scanner would have been more ideal.⁹ However, even in that study, the subjects had to walk over to the PET scanner after the MRI scan and had to undergo a second resting PET scan for comparison on a different day. The advent of newer MR-PET scanners offers an excellent opportunity to perform concurrent ASL/PET studies in order to more precisely map the correlations between these two modalities.^{2, 31}

Conclusion

Resting state correlations between cerebral perfusion as measured by pCASL and cerebral glucose metabolism as measured by ¹⁸F-FDG PET in this study were overall strong in whole-brain analysis, but there were considerable regional differences in this correlation. The striatum showed the highest correlation between perfusion and metabolism both across voxels and across subjects, whereas correlations were the weakest in medial temporal structures. In addition, rCMR_Glc was significantly greater than rCBF in the putamen, orbitofrontal and temporal lobes, whereas rCBF was higher in the hippocampus and insula. Although previous studies have looked at perfusion and metabolism in either an activating task or in a diseased state, ours is the first study with a relatively larger number of subjects to show regional correlations in healthy resting brains.

The authors declare no conflict of interest.