TL;DR
This study reveals that fluctuations in brain activity amplitude, often overlooked, are a significant source of variability in resting-state fMRI (rfMRI) data, both within and between individuals. These amplitude variations, particularly in sensory and motor networks, are linked to arousal levels and can drive observed changes in functional connectivity, suggesting a need to account for them in future analyses.
What Problem Does This Paper Address?
Interpreting resting-state fMRI (rfMRI) data, especially for clinical applications or individual-level analysis, hinges on understanding the sources of variability. While functional connectivity (FC) is widely studied, the spontaneous fluctuations in the BOLD signal's amplitude itself, and how these vary across individuals and over time within an individual, have been less explored. This variability in amplitude can confound FC estimates, making it difficult to distinguish true neural signals from noise or state-dependent changes. The authors aim to characterize this amplitude variability, understand its relationship with other factors like head motion and behavior, and determine its impact on FC measures.
Methods
The study employed two large-scale datasets (Human Connectome Project - HCP and UK Biobank) to investigate amplitude variations. Key methods include:
- Amplitude Estimation: Temporal standard deviation of the BOLD signal time series was calculated for each identified resting-state network (RSN). This is a direct measure of signal fluctuation magnitude.
- Clustering Analysis: Ward's clustering was used on the between-subject correlation matrix of RSN amplitudes to identify groups of networks whose amplitude variations covary similarly across individuals.
- Canonical Correlation Analysis (CCA): CCA was used to identify modes of covariation between RSN amplitudes and behavioral/lifestyle measures, and also between RSN amplitudes and head motion.
- Within-Subject Analysis: The four resting-state scans from the HCP dataset were used to examine how RSN amplitudes change within individuals over time, both across different scans and within individual scans.
- Additive Signal Change (ASC) Analysis: This novel approach was used to test whether observed changes in functional connectivity could be fully explained by changes in signal amplitude.
Datasets
- Human Connectome Project (HCP): Resting-state fMRI data from 819 subjects (S900 release). Each subject had 4 resting-state scans of 15 minutes each, acquired over two days.
- UK Biobank: Resting-state fMRI data from 5847 subjects. Each subject had one resting-state scan of 6.10 minutes. This dataset was used primarily for replication and to investigate correlations with arousal-related measures.
Key Insights and Ideas
- Sensory/Motor vs. Cognitive Networks: A clear distinction emerges in amplitude variability: primary sensory and motor networks show correlated amplitude fluctuations across subjects, separate from cognitive networks (like the Default Mode Network - DMN). This suggests fundamental differences in how these network types fluctuate.
- Within-Scan Amplitude Increase: Primary sensory/motor networks consistently show an increase in amplitude from the beginning to the end of a resting-state scan. This is a novel finding suggesting a dynamic change in signal magnitude within a single scanning session.
- Arousal as a Driver: The increase in sensory/motor amplitude over time, and higher amplitudes generally, are linked to lower arousal (e.g., sleepiness). This is supported by correlations with sleep duration in the UK Biobank data.
- Amplitude Drives Connectivity: Changes in functional connectivity between networks can be fully explained by changes in signal amplitude, rather than necessarily reflecting changes in the underlying neural coupling. This has significant implications for interpreting dynamic FC measures.
- Motion is Complex: While head motion is correlated with amplitude, the study provides evidence that the observed within-scan amplitude increases are not purely artifactual, suggesting motion might interact with or reflect underlying physiological states.
Key Results
- Clustering: Two distinct clusters of RSNs were identified based on amplitude covariation: a "primary sensory/motor cluster" (14 networks) and a "cognitive cluster" (11 networks).
- Within-Scan Changes: Primary sensory/motor networks showed a consistent increase in amplitude from the start to the end of scans, and also between scan 1 and scan 2 on the same day. Cognitive networks showed less consistent changes.
- Functional Connectivity: Changes in functional connectivity between scan 1 and scan 2 on the same day were widespread and could be fully explained by additive signal changes (i.e., amplitude changes).
- Arousal Link: A significant negative correlation was found between primary sensory/motor amplitude and sleep duration in the UK Biobank data (r = -0.11, p = 8*10–16).
- Motion Correlation: Between-subject correlations showed a link between primary sensory/motor amplitude and motion, though this was not replicated in the UK Biobank data. Within-subject, primary sensory/motor amplitude was significantly correlated with motion.
Limitations and Future Work
- Arousal Measurement: The study relies on self-report measures for arousal (sleep duration), which may not be perfectly accurate. Future work could benefit from more objective measures of arousal.
- Artifact vs. Signal: While the study provides strong evidence against amplitude changes being purely artifactual, disentangling the precise contribution of motion and physiological noise from genuine neural signals remains a challenge.
- Temporal Resolution: The findings regarding within-scan amplitude changes might be more pronounced in protocols with longer scan times. However, significant effects were observed even in shorter segments, suggesting this is a general phenomenon.
- Future Directions: The authors suggest that future rfMRI studies should consider obtaining arousal-related measures as covariates and explicitly investigate the influence of amplitude changes on functional connectivity, especially dynamic measures. Further research into the physiological underpinnings of these amplitude fluctuations is also warranted.
Summary updated