Resting-state BOLD local synchrony as a strong proxy of glucose uptake and as a biomarker of aging using functionally-driven gray matter parcelization
Michaël Bernier1, Étienne Croteau2, Christian-Alexandre Castellano2, Stephen C Cunnane2, and Kevin Whittingstall3

1Nuclear medecine and radiobiology, Université de Sherbrooke, Sherbrooke, QC, Canada, 2Research center of aging, Université de Sherbrooke, Sherbrooke, QC, Canada, 3Diagnostic radiology, Université de Sherbrooke, Sherbrooke, QC, Canada


Currently, PET is the primary imaging modality used to infer energy metabolism in the brain. It is also known to be a reliable biomarker of aging and cognitive diseases. However, the cost and invasive nature of PET limits its use in basic research. There is therefore great interest in developing alternative less invasive approaches for estimating brain glucose metabolism. Using resting state fMRI metrics such as regional local homogeneity (ReHo), amplitude of low-frequencies fluctuations (ALFF) and regional global connectivity (closeness) we found that both regional- and subject-variations in ReHo strongly correlate with brain glucose uptake in healthy young and aging participants.


Proper glucose metabolism (CMRGlu), approximated by [18F]-fluorodeoxyglucose positron emission tomography (PET FDG), is essential for maintaining synaptic activity in the brain. Synaptic activity increases brain glucose and oxygen metabolism which is then reflected by CBF increases1. The fMRI blood-oxygen-level dependent (BOLD) signal is related to CBF, CBV and oxygen metabolism in a complex fashion, so we sought to investigate its relationship with CMRGlu to understand the relation between BOLD and brain glucose consumption. We employed the best analytical approach (voxel vs structural or functional region based-analysis) to qualify fMRI measures as proxies of CMRGlu or biomarkers of aging.


We collected MRI and PET-FDG datasets in cognitively normal young (N=25, 18-30 years old;) and older (N=32, 65-85 years old) healthy participants. Each session started with an anatomical T1-weighted 1 mm isotropic MPRAGE (TR/TE 1860/3.54 msec) acquisition, followed by a resting-state fMRI dataset of 5 min using a standard echoplanar imaging (EPI) sequence (35 axial image slices, 64 x 64 matrix, TR/TE 2730/40 msec, voxel size 3.438 x 3.438 x 4.2 mm). Brain PET scans were then performed on a Philips Gemini TF PET/CT scanner (voxel size of 2 x 2 x 2 mm, field of view of 250 x 250 x 180 mm, 60 min with time frames of 12 x 10 s, 8 x 30 s, 6 x 4 min, 3 x 10 min). CMRGlu (μmol/100 g/min) was computed using PMOD 3.3 in the Patlak graphical model2 (lumped constant = 0.80) and defined as the product of the tracer uptake rate constant (K) multiplied by the plasma concentration of glucose. All fMRI analysis were carried out using a standard pipeline consisting of slice timing and motion correction and band-pass temporal filtering (0.005 to 0.1 Hz). Gaussian spatial smoothing was replaced by non-local mean denoising to improve signal SNR in subcortical areas3. Using AFNI4, we measured the amplitude of low frequencies fluctuations (ALFF)5 in the 0.01 – 0.1 Hz range as well as the regional local homogeneity (ReHo), the time series application of Kendall's W coefficient of concordance; For each voxel, a score of homogeneity between 27 neighbourhood voxels is computed in which a higher score reflects higher BOLD synchrony6. We also investigated the regional global homogeneity (closeness) computed as the sum of all correlations between parcellated regions. CMRGlu maps and voxel-based measures of ALFF and ReHo were registered to MNI standard space using ANTs7, allowing a direct comparison between FDG and fMRI datasets. For this, we parcellated the gray matter (GM) using Ward feature agglomeration clustering8 first based on the CMRGlu (50, 200 and 1000 clusters, respectively), then based on ReHo (50, 200 and 1000 clusters). For comparison, we subdivided the GM into 42 Freesurfer regions9 and no clustering (i.e. all voxels in the brain were pooled together).


For both young and older groups, Fig. 1 shows a scatter plot of CMRGlu vs (a) ALFF (p >0.05), (b) ReHo (p<0.05) and (c) closeness (p<0.05) when using a 200 CMRGlu-based Ward clustering (results were similar using clusters of 50 and 1000 thus not shown here). Fig. 2 shows how different parcelization strategies affect the ReHo-CMRGlu relation. Although the correlation values change, ReHo is consistently better correlated to CMRGlu compared to ALFF, as shown on Fig. 3. Finally, Fig. 4 illustrates the potential of ReHo as a successful biomarker for aging since it can easily differentiate the younger and older group as effectively as CMRGlu.

Discussion & Conclusion

Using different gray matter parcelization strategies, we consistently observed that the amplitude of the BOLD signal (ALFF) is a poor predictor of CMRGlu (Fig. 3) whereas local temporal synchrony (ReHo) is strongly correlated to CMRGlu, and thus a potentially promising proxy of glucose intake (Fig. 1-2, Y: R=0.85, A: R=0.81) and biomarker of aging (Fig.4). However, the correlation between ReHo and CMRGlu varied depending on the manner in which the GM is parcellated (see Fig. 2). The physiological meaning of the strong ReHo-CMRGlu relationship is unclear; It has been proposed that resting-state BOLD fluctuations might be correlated to vasodilatation caused by CBF, possibly revealed by the local connectivity10 (ReHo). However, the fact that long-range connectivity (closeness or networks) is also observed in the same signal implies that vasomotion is homogeneous throughout the vasculature, an assumption that still requires verification. Since CMRGlu was shown to be strongly correlated to CBF1, further investigation is needed to isolate the neurovascular underpinnings the local BOLD synchrony. Nevertheless, our data suggest that it can be used as an accurate marker for CMRGlu.


No acknowledgement found.


1 van Golen LW, Huisman MC, Ijzerman RG, Hoetjes NJ, Schwarte LA, Lammertsma AA et al. Cerebral blood flow and glucose metabolism measured with positron emission tomography are decreased in human type 1 diabetes. Diabetes 2013; 62: 2898–904.

2 Patlak CS, Blasberg RG, Fenstermacher JD. Graphical evaluation of blood-to-brain transfer constants from multiple-time uptake data. J Cereb Blood Flow Metab 1983; 3: 1–7.

3 Bernier M, Chamberland M, Houde J-C, Descoteaux M, Whittingstall K. Using fMRI non-local means denoising to uncover activation in sub-cortical structures at 1.5 T for guided HARDI tractography. Front Hum Neurosci 2014; 8: 715.

4 Cox RW. AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res an Int J 1996; 29: 162–173.

5 Zuo X-N, Di Martino A, Kelly C, Shehzad ZEZE, Gee DGDG, Klein DFDF et al. The oscillating brain: complex and reliable. Neuroimage 2010; 49: 1432–45.

6 Zang Y, Jiang T, Lu Y, He Y, Tian L. Regional homogeneity approach to fMRI data analysis. Neuroimage 2004; 22: 394–400.

7 Avants B, Tustison N, Song G. Advanced Normalization Tools (ANTS). Insight J 2009; : 1–35.

8 Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O et al. Scikit-learn: Mchine Learning in Python. J Mach Learn Res 2011; 12: 2825–2830.

9 Desikan RS, Ségonne F, Fischl B, Quinn BT, Dickerson BC, Blacker D et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage 2006; 31: 968–80.

10 Lu H, Stein EA. Resting state functional connectivity: its physiological basis and application in neuropharmacology. Neuropharmacology 2014; 84: 79–89.


Using a Ward CMRGlu based clustering (200 grey matter clusters), we show for both the young and older groups the correlation between a) ALFF (Y & A: R=0, p>0.05), b) Closeness (Y & A: R=0.512, p<0.005) and c) ReHo (Y: R=0.860, A: R=0.814, p < 0.005). Each dot represents the mean of all subject for a cluster. A strong correlation was observed between ReHo and CMRGlu.

For both young and aging groups, the relation between ReHo and CRMGlu in the grey matter is illustrated with the mean of all subjects for multiple parcelization techniques (p<0.005): a) All voxels (117000 clusters, Y: R=0.43, A: R=0.39), b) Freesurfer (42 clusters, Y: R=0.61, A: R=0.62), c) CMRGlu Ward (200 clusters, Y: R=0.86, A: R=0.81), d) Ward (200 clusters, Y: R=0.69, A: R=0.68), e) CMRGlu Ward (1000 clusters, Y: R=0.86, A: R=0.79), f) Ward (1000 clusters, Y: R=0.48, A: R=0.50).

We compared the a-b) ReHo-CMRGlu and the c-d) ALFF-CMRGlu relation using either a b-d) region-based (Ward CMRGlu based clustering with 200 clusters) or a a-c) voxel-based approach. No significant correlation was found in ALFF compared to ReHo.

Using a Ward CMRGlu based clustering (50 clusters), we computed the difference between the mean of all young and the mean of all aging for every clusters. All ROIs are ordered in term of size (1 being the largest). A positive value means the value was higher in the younger population; Both CMRGlu and ReHo were significantly stronger in the young population for most of the clusters. ReHo performed as well as CMRGlu to differentiate populations based on age.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)