Reliable and robust causal inference approaches for effective connectivity research with fMRI data

NIH RePORTER · NIH · P20 · $402,724 · view on reporter.nih.gov ↗

Abstract

Neuroscientists grapple with understanding the causal mechanisms within brain networks that govern perception, cognitive functions, decision-making, and behavior. To understand these mechanisms, researchers have attempted to translate statistical associations between brain regions (so-called functional connectivity) into causal relationships, raising the question of whether one brain region has a direct influence on the physiological activity recorded in other brain regions. Such causal relationships are called effective connectivity. Effective connectivity is essential in learning how neural activities in different regions are causally related. It also provides important evidence for designing future experiments aiming at affecting certain cognitive outcomes by intervening on specific neural processes. Several methods have been developed for identifying and estimating effective connectivity, including Granger causality and dynamic causal modeling. However, these existing methods are vulnerable to spurious associations due to the shared network, temporal, and spatial dependence structures found in neuroimaging data. Moreover, they are not often explicit about the assumptions on unmeasured confounders, such as whether and how much unobserved neural activities that affect multiple brain regions are allowed when determining effective connectivity. These two common sources of spurious and biased findings can readily mislead our understanding of effective connectivity, resulting in poorly designed experiments or interventions for improved cognitive functions. The goal of this proposal is to develop reliable and robust causal inference methods to infer effective connectivity between brain regions that account for the shared dependence structures as well as unmeasured confounding factors. This proposed research will raise awareness of potential sources of bias and misleading findings in neuroimaging (e.g., fMRI) data, as well as provide more reliable and robust inferential tools than existing methods that are often relying on a single p-value from a single parameter in the presumed parametric model. This pilot research will pave the way for future independent funding that will further investigate effective connectivity among multiple brain regions that are robust to many sources of spurious findings.

Key facts

NIH application ID
10709066
Project number
5P20GM103645-10
Recipient
BROWN UNIVERSITY
Principal Investigator
Youjin Lee
Activity code
P20
Funding institute
NIH
Fiscal year
2022
Award amount
$402,724
Award type
5
Project period
2022-08-01 → 2025-07-31