Neural Event Time Series Based Functional Connectivity Analysis for Functional MRI
Abstract
Purpose
Functional connectivity (FC) derived from functional magnetic resonance imaging (fMRI) is a key marker of brain health and cognitive function. Most analyses rely on the blood-oxygen-level-dependent (BOLD) signal; however, BOLD reflects both neural activity and the hemodynamic response to stimuli. The latter differs between individuals, potentially confounding BOLD-based analyses and obscuring interregional neural relationships. This work evaluated a deep learning-based method for dissociating the hemodynamic response from the corresponding binary neural event time series (NETS) by comparing NETS-based FC patterns in healthier and less-healthy middle-aged individuals.
Methods
NETS were extracted from BOLD time series acquired during a Stroop task in sixty participants across fifteen brain regions. Fourteen regions were task-activated, while one uninvolved region served as a negative control. FC among all possible region pairs was analyzed. Mutual information (MI) was computed using both NETS and continuous BOLD signals, and Pearson correlation (PC) was computed using BOLD signals. Participants were stratified into low and high diastolic blood pressure groups. For each FC metric, Student’s t-tests and robust Cohen’s D effect size estimates were used to compare task-activated region pairs against the average FC between the negative control region and task-activated regions.
Results
All FC metrics showed greater connectivity in the healthier group than the less-healthy group. However, NETS-based analyses revealed unique connections not detected using BOLD-based comparisons, suggesting that removing hemodynamic effects allows access to distinct functional information. MI-NETS produced connectivity measures that differed from MI-BOLD, demonstrating the impact of hemodynamic information embedded in the BOLD signal.
Conclusion
FC estimates derived from BOLD fMRI are influenced by hemodynamic effects that may obscure inter-regional neural relationships. Binary NETS mitigates these confounds, enabling FC assessments that more directly reflect neural activity. This approach may be particularly valuable for studying brain function in populations with cardiovascular conditions that alter cerebral blood flow.