Abstract
Functional MRI (fMRI) studies reported disruption of resting-state networks (RSNs) in several neuropsychiatric disorders. PET with $^{18}$F-FDG captures neuronal activity that is in steady state at a longer time span and is less dependent on neurovascular coupling. In the present study, we aimed to identify RSNs in $^{18}$F-FDG PET data and compare their spatial pattern with those obtained from simultaneously acquired resting-state fMRI data in 22 middle-aged healthy subjects. Thirteen and 17 meaningful RSNs could be identified in PET and fMRI data, respectively. Spatial overlap was fair to moderate for the default mode, left central executive, primary and secondary visual, sensorimotor, cerebellar, and auditory networks. Despite recording different aspects of neural activity, similar RSNs were detected by both imaging modalities. The results argue for the common neural substrate of RSNs and encourage testing of the clinical utility of resting-state connectivity in PET data.
Introduction
Mapping the brain’s resting-state networks (RSNs)—coherent patterns of activity occurring in the absence of specific tasks—has traditionally relied on functional MRI (fMRI). However, fMRI measures the BOLD (blood-oxygen-level-dependent) signal, an indirect proxy for neural activity influenced by vascular health and hemodynamics. In contrast, Positron Emission Tomography (PET) using $^{18}$F-FDG directly measures glucose metabolism, providing a stable indicator of neuronal energy consumption. This study utilizes simultaneous PET/fMRI acquisition to investigate whether these fundamentally different imaging signals converge on the same anatomical networks within the same subjects at the same moment.
Methods
The study involved 22 healthy, right-handed subjects (mean age 54.5 years). Data were acquired using a hybrid PET/MR scanner, recording BOLD fluctuations and glucose metabolism concurrently over a 10-minute resting period. Independent Component Analysis (ICA) was applied to both modalities: temporal ICA for fMRI to identify voxels fluctuating together in time, and intersubject covariance ICA for PET to identify regions where glucose metabolism intensity varied synchronously across the group. Spatial similarity between the resulting networks was quantified using the Dice similarity coefficient.
Results and Discussion
The analysis identified 13 meaningful RSNs in PET and 17 in fMRI. Ten networks were common to both modalities, including the Default Mode Network (DMN), primary visual, sensorimotor, and auditory networks.
As shown in Figure 1, spatial overlap was "fair to moderate," with the highest similarity observed in the DMN and primary visual cortex. These findings suggest that RSNs represent fundamental biological units rather than imaging artifacts. However, the modalities also exhibited unique sensitivities. PET was more effective at identifying subcortical networks (e.g., basal ganglia and thalamus), while fMRI excelled at capturing higher-order cognitive networks like the salience and spatial attention networks.
The study's limitations include a relatively small sample size for intersubject covariance analysis and the use of static PET data. Future work utilizing dynamic PET (fPET) may further improve spatial similarity by enabling temporal correlation analysis.
Conclusion
The convergence of fMRI and PET signals on a shared functional architecture confirms a common neural substrate for resting-state networks. Given that PET is a clinical staple for neurodegenerative diseases, these results support the potential for "metabolic connectivity" as a reliable biomarker for assessing functional brain health in clinical settings.
Summary updated