Estimating Task‐Evoked Neurovascular Coupling Using Mutual Information Between <scp>BOLD</scp> and Perfusion‐Weighted <scp>fMRI</scp> Signals

Understanding neurovascular coupling (NVC) is essential for interpreting functional MRI (fMRI) data, particularly in task-based paradigms. BOLD-ASL coupling studies, where simultaneously collected blood oxygenation dependent (BOLD) and arterial spin labeling (ASL) signals are correlated, have shown widespread coupling during the resting state. While prior work has demonstrated widespread BOLD-ASL coupling at rest, less is known about how this coupling behaves during tasks and whether nonlinear dependencies contribute to NVC. This study investigates BOLD-ASL coupling during a visual checkerboard task and a finger-tapping motor task. Coupling was evaluated at zero lag and at the lag of maximum dependence. Spatial correspondence with task activation was quantified using Dice coefficients within Yeo 17-network regions. Coupling was evaluated using traditional Pearson correlation (corr) and mutual information (MI), a model-free approach capable of detecting both linear and nonlinear dependencies. Results showed strong spatial correspondence between task activation and BOLD-ASL coupling for both tasks, particularly in expected visual, sensory, and motor regions. Notably, MI showed more overlap with traditional GLM-based task activation models compared to corr. This suggests that MI may provide additional insights into the complexity of NVC beyond traditional linear methods. These findings reinforce the importance of using multimodal fMRI approaches to characterize NVC more comprehensively.