Most of what we know about the human brain in action comes from a proxy. Functional Magnetic Resonance Imaging (fMRI) does not measure the electrical firing of neurons directly; instead, it tracks the "Blood-Oxygen-Level-Dependent" (BOLD) signal. This signal relies on neurovascular coupling (NVC)—the process by which active neurons signal local blood vessels to dilate, bringing in fresh oxygenated hemoglobin.
While this proxy is powerful, it creates a fundamental ambiguity: when a signal changes, is it because the neurons changed their firing, or because the vascular "plumbing" responded differently? Researchers are increasingly focused on disentangling these two drivers to ensure that our maps of the mind aren't actually maps of the circulatory system.
The Cellular Messengers of Blood Flow
To understand why the BOLD signal works, we must first identify the bridge between synapses and vessels. In the cerebellum, Josee M.T. Huard, Costantino Iadecola, and colleagues demonstrated that this isn't a generic property of all cells. In Stellate Neurons Mediate Functional Hyperemia in the Cerebellar Molecular Layer, they used genetic knockout models to show that stellate neurons act as the essential interneurons that translate synaptic activity into vascular dilation.
This finding provides the biological "ground truth" for the linear transform model proposed by Geoffrey M. Boynton and David Heeger. In their foundational work, Linear Systems Analysis of Functional Magnetic Resonance Imaging in Human V1, they showed that despite its sluggishness, the hemodynamic response behaves predictably enough to infer underlying neural properties like contrast sensitivity.
The Global Signal as Autonomic Choreography
For years, researchers treated the "Global Signal"—widespread fluctuations across the entire brain—as a nuisance to be scrubbed away. However, recent evidence suggests this signal is a window into the brain-body axis. Pınar S. Özbay and Peter van Gelderen found that systemic sympathetic bursts during sleep trigger brain-wide vasoconstriction, directly dropping the fMRI signal (Sympathetic activity contributes to the fMRI signal).
This suggests that large-scale brain patterns aren't always "thinking" patterns. Taylor Bolt and Jason S. Nomi have shown that these fluctuations are tightly coupled with respiration, heart rate, and pupil dilation (Widespread Autonomic Physiological Coupling Across the Brain-Body Axis). If we mistake these autonomic rhythms for neural connectivity, we risk misinterpreting how regions communicate.
The Aging Confound: Stiff Pipes vs. Quiet Neurons
The challenge of disentangling these signals is most acute in aging and disease. As we age, our cerebral vasculature stiffens and blood flow dynamics change. Lyuan Xu and Kurt G. Schilling have pioneered the study of functional signals in white matter, the brain's wiring (Functional correlation tensors in brain white matter and the effects of normal aging).
They found that white matter BOLD signals—once dismissed as noise—show clear age-related declines. However, a persistent question remains: does a weaker signal in an 80-year-old brain mean the neurons are less active, or simply that the aging vessels can no longer dilate efficiently? Muwei Li notes that hemodynamic measures struggle to differentiate between metabolic demand and vascular health, making it difficult to isolate purely neurological degradation (Changes in white matter functional networks across late adulthood).
Validating the Proxy
To trust hemodynamic imaging, we must ground it in direct electrical measurements. Michael Krumin and Matteo Carandini achieved this by combining functional ultrasound (fUSI) with high-density Neuropixels recordings (Neural correlates of blood flow measured by ultrasound). They proved that blood flow is a faithful, linear reflection of slow neural firing rates, particularly from inhibitory populations.
Similarly, Aaron Kucyi used intracranial electrophysiology in surgical patients to confirm that the "intrinsic networks" seen in fMRI are indeed rooted in reproducible patterns of high-frequency neural activity (Intracranial Electrophysiology Reveals Reproducible Intrinsic Functional Connectivity within Human Brain Networks).
Toward Multimodal Fusion
The future of the field lies in mathematical frameworks that don't just "clean" the data, but model the interaction between neurons and vessels. Karl J. Friston and Adeel Razi have proposed Bayesian fusion, which uses the high-speed timing of EEG to set "priors" for fMRI (Bayesian fusion and multimodal DCM for EEG and fMRI). By modeling the conditional dependence of blood flow on neural firing, these tools allow researchers to estimate connectivity with a precision that neither modality could achieve alone.
By treating the vascular system not as a source of noise, but as a structured biological system with its own rules, we can finally see through the "blood" to the "brain" beneath.