TL;DR
This paper introduces the use of fMRI phase data, in addition to the standard magnitude data, to study brain functional connectivity. They found that phase-based connectivity analysis reveals a sparser and more balanced network structure compared to magnitude-based analysis, suggesting phase data offers a more direct representation of the brain's internal magnetic state.
What Problem Does This Paper Address?
Functional Magnetic Resonance Imaging (fMRI) is a cornerstone for understanding brain activity. Conventionally, researchers analyze the magnitude of the fMRI signal, which reflects the overall strength of the magnetic resonance signal. However, fMRI data is inherently complex-valued, consisting of both magnitude and phase components. The phase component is directly related to the local magnetic field variations within the brain. While this phase information is generated alongside magnitude data at no extra cost, it has been less explored for functional connectivity analysis. This paper aims to bridge this gap by investigating whether the phase component of fMRI data provides unique or complementary insights into brain functional connectivity, particularly in the resting state, and how it compares to the established magnitude-based approach.
Methods
The study employed a large cohort of 600 participants to analyze resting-state fMRI data. The core methodology involved:
- Data Acquisition: Simultaneous acquisition of both magnitude and phase fMRI data.
- BOLD-Only Phase Extraction: A novel method was used to extract the "BOLD-only" phase signal, which is thought to be more directly related to neuronal activity by removing static field contributions. This involved complex division and a perturbation model to isolate dynamic phase changes.
- Group Independent Component Analysis (ICA): Both magnitude (mICA) and phase (pICA) data were decomposed into independent components, representing distinct brain subnetworks. A group-informed ICA approach (GIG-ICA) was used for phase data to ensure correspondence with the magnitude components.
- Functional Network Connectivity (FNC) Calculation: Temporal correlations between the ICA timecourses were computed to generate FNC matrices for both magnitude (mFNC) and phase (pFNC).
- Statistical Analysis: Null-hypothesis tests (one-sample t-tests) were performed on the group FNC matrices to identify statistically significant connections. False Discovery Rate (FDR) correction was applied to control for multiple comparisons.
- Characterization of Connectivity: The study analyzed various characteristics of the FNC matrices, including:
- Sparsity: The number of significant connections.
- Balance: The distribution of positive and negative correlations.
- Intra- vs. Inter-domain connections: Distinguishing between connections within a brain region (near) and between different regions (far).
- Strength and Robustness: Assessing the significance and consistency of connections across different p-value thresholds.
Datasets
The study utilized a dataset of 600 participants (age 10-76, 346 males) from the Mind Research Network. This dataset included pairs of resting-state fMRI magnitude and phase images. No new benchmark datasets were introduced; rather, a large existing dataset was used to compare the two types of fMRI data.
Key Insights and Ideas
- Phase Data Reveals a "Sparser" Brain Network: The most striking finding is that phase-based FNC (pFNC) consistently shows fewer significant connections than magnitude-based FNC (mFNC), especially at stringent statistical thresholds. This suggests that the phase signal might be more sensitive to specific, perhaps more fundamental, interactions.
- Phase Data Shows Better "Balance": The pFNC matrices exhibit a more balanced distribution of positive and negative functional connections compared to mFNC, which tends to be dominated by positive correlations. This balance is attributed to the bipolar nature of magnetic fields and the fact that phase data preserves negative correlations, whereas magnitude data can suppress them.
- Direct Representation of Magnetic State: The paper emphasizes that phase fMRI directly reflects the brain's internal magnetic fieldmap, offering a more direct representation of the brain's magnetic state compared to magnitude, which is an indirect measure.
- Subcortical "Clique" Identified: Phase analysis highlighted strong intra-domain couplings within subcortical nuclei (basal ganglia and thalamus), forming a functional "clique" that is negatively coupled with visual and cerebellar regions. This finding aligns with existing knowledge of the subcortex's role as a central hub.
Key Results
- At a p-value threshold of $p < 10^{-200}$, mFNC showed 60 positive and 1 negative connection, while pFNC showed 10 positive and 2 negative connections.
- The mean of the mFNC matrix was consistently higher than the mean of the pFNC matrix across various p-value thresholds, indicating a greater imbalance in magnitude-based connectivity.
- The average absolute strength of connections (mean(|FNC|)) was often higher in pFNC than mFNC, suggesting that the significant connections identified in phase data might be stronger.
- The study found that phase data reveals a more balanced distribution of positive and negative connections, with smaller differences between the counts and sums of positive and negative connections compared to magnitude data.
Limitations and Future Work
- "Ground Truth" Uncertainty: The authors acknowledge that without a definitive "ground truth" for in vivo brain function connectivity, it's challenging to definitively prove the superiority of phase data.
- Linear Approximation: While the methods aim to linearize phase signals, the underlying physics of fMRI phase is complex, and the linear approximation might not hold perfectly in all scenarios.
- Future Work: The paper suggests further research into more accurate inverse mapping of phase to magnetic fieldmap and exploring the utility of phase fMRI in task-based paradigms and clinical applications. The potential for phase fMRI to provide a more robust and less artifact-prone measure of functional connectivity is a key avenue for future exploration.
Summary updated