Abstract:Granger Causality (GC) is widely used in neuroimaging to estimate directed statistical dependence among brain regions using time series of brain activity. An important issue is that functional MRI (fMRI) measures brain activity indirectly via the blood-oxygen-level-dependent (BOLD) signal, which affects the temporal structure of the signals and distorts GC estimates. However, some notable applications of GC are not concerned with the GC magnitude but its statistical significance. This is the case for network inference, which aims to build a statistical model of the system based on directed relationships among its elements. The critical question for the viability of network inference in fMRI is whether the hemodynamic response function (HRF) and its variability across brain regions introduce spurious relationships, i.e., statistically significant GC values between BOLD signals, even if the GC between the neuronal signals is zero. It has been mathematically proven that such spurious statistical relationships are not induced if the HRF is minimum-phase, i.e., if both the HRF and its inverse are stable (producing finite responses to finite inputs). However, whether the HRF is minimum-phase has remained contentious. Here, we address this issue using multiple realistic biophysical models from the literature and studying their transfer functions. We find that these models are minimum-phase for a wide range of physiologically plausible parameter values. Therefore, statistical testing of GC is plausible even if the HRF varies across brain regions, with the following limitations. First, the minimum-phase condition is violated for parameter combinations that generate an initial dip in the HRF, confirming a previous mathematical proof. Second, the slow sampling of the BOLD signal (in seconds) compared to the timescales of neural signal propagation (milliseconds) may still introduce spurious GC.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Whether the hemodynamic response function (HRF) is a minimum - phase system and what impact this has on Granger Causality (GC) analysis in functional magnetic resonance imaging (fMRI)**. Specifically, the paper focuses on the influence of the minimum - phase property of HRF on GC estimation. HRF describes the temporal change of the blood - oxygen - level - dependent (BOLD) signal induced by brain neural activity. Since fMRI indirectly measures brain activity through the BOLD signal, and the temporal structure of the BOLD signal is affected by HRF, this may lead to distortion in GC estimation. In particular, when HRF is not minimum - phase, false statistical relationships may be introduced, that is, in the absence of a causal relationship between neural signals, significant GC values are shown between BOLD signals. ### Main problems: 1. **Whether HRF is a minimum - phase system**: If HRF is minimum - phase, then its inverse system is also stable, which can ensure that no false causal relationships are introduced. 2. **The influence of HRF changes on network inference**: When performing network inference in fMRI, whether changes in HRF will introduce false GC relationships, thereby affecting the accuracy of the network model. ### Research methods: - The paper studies the transfer functions of two commonly used biophysical HRF models (Stephan et al., 2007 and Havlicek et al., 2015) by linearizing them and determines the minimum - phase conditions of these models under different parameters. - At the same time, it also studies the minimum - phase property of the canonical HRF model used in SPM12 software. ### Conclusions: - **Stephan et al. (2007) model**: Within a reasonable physiological parameter range, this model satisfies the minimum - phase condition, unless the parameter combination causes an initial dip in HRF. - **Havlicek et al. (2015) model**: Within a reasonable parameter range, this model also satisfies the minimum - phase condition. - **Canonical HRF model**: Within a wide range of parameters, including the SPM12 default parameters, this model also satisfies the minimum - phase condition. ### Practical significance: - If HRF is minimum - phase, then even if HRF varies in different brain regions, GC statistical tests based on BOLD signals are still feasible, although there are still some limitations, such as slow sampling rates may introduce false GC. In summary, through mathematical proofs and model analysis, this paper explores the influence of the minimum - phase property of HRF on GC analysis in fMRI, providing important insights into the temporal characteristics of BOLD signals and their impact on causal relationship inference.

Minimum-phase property of the hemodynamic response function, and implications for Granger Causality in fMRI

Is Granger Causality a Viable Technique for Analyzing fMRI Data?

Global Fluctuations of Cerebral Blood Flow Indicate a Global Brain Network Independent of Systemic Factors.

Directed neural interactions in fMRI: a comparison between Granger Causality and Effective Connectivity

Causality analysis of neural connectivity: critical examination of existing methods and advances of new methods.

New Insights into Signed Path Coefficient Granger Causality Analysis

The effect of calcium restriction in the diet of calcium transport in rat small intestine.

Detectability of Granger causality for subsampled continuous-time neurophysiological processes

Spatio-temporal Granger causality: A new framework.

An information-theoretic framework for conditional causality analysis of brain networks

Change in Human Brain Functional Network Based on Granger Causality Analysis

Multivariate Granger causality unveils directed parietal to prefrontal cortex connectivity during task-free MRI

Benchmarking nonparametric Granger causality: Robustness against downsampling and influence of spectral decomposition parameters

Evaluating the effective connectivity of resting state networks using conditional Granger causality

Granger causal influence predicts BOLD activity levels in the default mode network

Linking brain vascular physiology to hemodynamic response in ultra-high field MRI

A Conditional Granger Causality Model Approach for Group Analysis in Functional Magnetic Resonance Imaging

A Stepwise Multivariate Granger Causality Method for Constructing Hierarchical Directed Brain Functional Network

Continuum - removed absorption features estimate tropical savanna grass quality in situ

Directed networks and resting-state effective brain connectivity with state-space reconstruction using reservoir computing causality

Unified Model Selection Approach Based on Minimum Description Length Principle in Granger Causality Analysis.