Biophysical simulations of fMRI responses using realistic microvascular models: Insights into distinct hemodynamics...

Functional magnetic resonance imaging (fMRI) is broadly used to measure human brain activity, however, the hemodynamic changes that comprise the fMRI response to neuronal activity are often interpreted using microscopy data in mice. These microscopy data provide ground-truth observations of how individual blood vessels respond to neuronal activity and thus form the basis of our fundamental understanding of neurovascular coupling. Although these invasive experiments provide invaluable insight, there are striking differences in the vascular architecture of mouse and human brains that may influence the hemodynamic response. Motivated by this, we developed a biophysical modeling framework for realistic hemodynamic simulations in both mouse and human cerebral cortex. For this, we utilized Vascular Anatomical Network (VAN) models that explicitly represent the full microvascular tree as a single connected network, originally based on anatomical reconstructions from a given location of mouse cerebral cortex. We extended the VAN modeling framework using synthetic VAN models representing the microvascular network at a single location of the human cerebral cortex. To account for larger size and complexity of the human VAN models, we developed an efficient computational framework to simulate the full hemodynamic responses in this human model and compared the simulated fMRI responses between mice and humans. Our biophysical simulations are based entirely on first principles (e.g., conservation of mass); model parameter values were fixed across all simulations, not tuned to fit data, as they represent meaningful physical constants taken from previous measurements. Only two simple calibrations were tuned for each simulation, to match baseline perfusion rates (blood flow) and oxygen extraction (OEF). Our results show that differences in microvasculature indeed influenced the hemodynamic response and led to observable differences in timing-for example, the simulated fMRI response peak in humans was delayed by ~2 s compared with that of mice, consistent with prior fMRI observations. While there are many known differences in vascular architecture in rodents and humans, we also discovered that, unexpectedly, an asymmetry in the number of branches of the penetrating intracortical arterioles and venules appears to be conserved across species. We demonstrate through further simulations that this anatomical property may also be needed for suitable hemodynamic responses. Our framework thus provides a valuable tool for bridging in-vivo microscopy of microvascular dynamics to human fMRI.