Biophysical Modelling to Deconvolve Neurovascular Signals
Jingyuan Chen1
1MGH/HST Martinos Center, United States

Synopsis

Functional magnetic resonance imaging (fMRI) tracks neuronal activities indirectly through neurovascular modulation of T2 or T2* relaxivity consequent to altered neuronal metabolism. Dissecting neurovascular dynamics driving the macroscopic fMRI fluctuations will not only lead to more accurate characterizations of neuronal activities, but also provide a wealth of meaningful biophysical metrics for clinical and neuroscience inferences. This lecture will first explain how various hemodynamic components responsive to neuronal activities (including oxygen metabolism, cerebral blood flow and blood volume) give rise to transient dynamics in fMRI signals; then provide an overview of models commonly employed to deconvolve neurovascular information.

Target Audience

Faculty and trainees with basic knowledge of blood-oxygenation-level-dependent (BOLD) fMRI that are interested in fMRI signal modeling.

Outcomes/Objectives

At the end of the lecture, the audience will understand (1) the biophysical relationship between different factors that drive hemodynamic changes; and (2) the rationales, applications and limitations of common neurovascular models.

Introduction

Since its inception in early 1990s, fMRI has been the dominant tool for non-invasively mapping brain function and physiology1-3. Yet, as an indirect measure of neural activity, BOLD contrast relies on a combination of changes in cerebral metabolic rate of oxygen (CMRO2), cerebral blood flow (CBF), and cerebral blood volume (CBV)4; alterations in any facets of these metrics relating to task, aging or pathology will elicit variability in fMRI signals despite unaltered underlying neural responses. Hence, deconvolving neuronal activities from all these “confounding” vascular factors is essential for improving the neuronal specificity of fMRI.

Several biophysical models have been proposed in the past two decades, building off human MR and rodent optical imaging data, and hence have opened the possibility to disentangle various pieces driving the hemodynamic changes and to unveil the underlying neuronal processes of interest. This lecture is aimed at providing an elementary overview of common models capturing the cascading metabolic and vascular changes giving rise to BOLD dynamics; and preparing the audience for the ensuing live demonstration on implementing these common models. The scope of this lecture will be limited to modeling the dynamic, transient features of neurally-driven hemodynamic changes, as steady-state conditions (including calibrated BOLD) are discussed in a separate educational talk.

Outline

The lecture will be organized into four sections, as detailed below:

(i) Neural and vascular drivers of hemodynamic changes. The first section will review existing theory and rationales on neurovascular/neurometabolic coupling, flow-volume (un)coupling, and the influence of baseline physiology exerted on the dynamic range of BOLD contrasts, accompanied by examples demonstrating how alterations of different vascular components can lead to notable hemodynamic changes through tasks or pharmacological challenges.

(ii) Modeling the hemodynamic response to brain activation. Here we describe two categories of approaches that are commonly employed to deconvolve neurovascular processes. The first category comprises a set of generative approaches that explicitly model the biophysical processes linking neural activities to BOLD signals. We begin by introducing a generic framework proposed in Buxton et al. (2004) that unifies the most common mathematical models involved in hemodynamic modeling4. We will then describe extensions of this model to incorporate neuronal and metabolic dynamics reflecting the excitatory-inhibitory balance5,6, and the spatial dependence of propagating hemodynamic waves7,8. The second category of approaches depend on presumed temporal patterns of the impulse hemodynamic response and additional regularization (e.g., spatial or temporal smoothness) to jointly characterize neuronal processes and hemodynamic signals9,10, i.e., devoid of explicit knowledge of the underlying neuro-/vascular-physiology.

(iii) Applications. This section will demonstrate the use of the aforementioned models with two examples: (1) inferring effective connectivity between pairs of cortical regions11; and (2) guiding simulations and interpretations of emerging fast fMRI signals12,13.

(iv) Limitations & Outlooks. Finally, we briefly highlight the limitations and topics of debate centered around these modeling endeavors, and discuss the need to adapt them for modern fMRI acquisitions with high spatiotemporal resolutions that were unavailable when the models were introduced.

Acknowledgements

Supported in part by the NIH NIBIB (grants P41-EB015896, and R01-EB019437), by the BRAIN Initiative (NIH NIMH grants R01-MH111438 and R01-MH111419), and by the MGH/HST Athinoula A. Martinos Center for Biomedical Imaging.

References

[1] Kwong et al. "Dynamic magnetic resonance imaging of human brain activity during primary sensory stimulation." Proceedings of the National Academy of Sciences 89.12 (1992): 5675-5679.

[2] Ozawa et al. "Intrinsic signal changes accompanying sensory stimulation: functional brain mapping with magnetic resonance imaging." Proceedings of the National Academy of Sciences 89.13 (1992): 5951-5955.

[3] Bandettini et al. "Time course EPI of human brain function during task activation." Magnetic resonance in medicine 25.2 (1992): 390-397.

[4] Buxton et al. "Modeling the hemodynamic response to brain activation." Neuroimage 23 (2004): S220-S233.

[5] Marreiros et al. "Dynamic causal modelling for fMRI: a two-state model." Neuroimage 39.1 (2008): 269-278.

[6] Havlicek et al. "Physiologically informed dynamic causal modeling of fMRI data." Neuroimage 122 (2015): 355-372.

[7] Aquino et al. "Deconvolution of neural dynamics from fMRI data using a spatiotemporal hemodynamic response function." Neuroimage 94 (2014): 203-215.

[8] Havlicek et al. "A dynamical model of the laminar BOLD response." Neuroimage 204 (2020): 116209.

[9] Karahanoğlu et al. "Total activation: fMRI deconvolution through spatio-temporal regularization." Neuroimage 73 (2013): 121-134.

[10] Wu et al. "A blind deconvolution approach to recover effective connectivity brain networks from resting state fMRI data." Medical image analysis 17.3 (2013): 365-374.

[11] Friston et al. "Dynamic causal modelling." Neuroimage 19.4 (2003): 1273-1302.

[12] Chen et al. "BOLD fractional contribution to resting-state functional connectivity above 0.1 Hz." Neuroimage 107 (2015): 207-218.

[13] Lewis et al. "Fast fMRI can detect oscillatory neural activity in humans." Proceedings of the national academy of sciences 113.43 (2016): E6679-E6685.

Proc. Intl. Soc. Mag. Reson. Med. 28 (2020)