[Paper Review] Towards Robust and Specific Causal Discovery from fMRI.

Kun Zhang, Biwei Huang+4 more|arXiv (Cornell University)|Sep 27, 2015

Functional Brain Connectivity Studies3 citations

TL;DR

This paper proposes a time- and condition-specific constraint-based causal discovery method for fMRI that treats time or experimental condition as an additional variable to improve robustness and specificity. By leveraging time-delayed conditional independence at neural activity frequencies, the method reduces spurious connections and enables accurate causal skeleton and direction identification, validated on simulated and real fMRI data with strong performance.

ABSTRACT

There are several issues with causal discovery from fMRI. First, the sampling frequency is so low that the time-delayed dependence between different regions is very small, making time-delayed causal relations weak and unreliable. Moreover, the complex correspondence between neural activity and the BOLD signal makes it difficult to formulate a causal model to represent the effect as a function of the cause. Second, the fMRI experiment may last a relatively long time period, during which the causal influences are likely to change along with certain unmeasured states (e.g., the attention) of the subject which can be written as a function of time, and ignoring the time-dependence will lead to spurious connections. Likewise, the causal influences may also vary as a function of the experimental condition (e.g., health, disease, and behavior). In this paper we aim to develop a principled framework for robust and time- or condition-specific causal discovery, by addressing the above issues. Motivated by a simplified fMRI generating process, we show that the time-delayed conditional independence relationships at the proper causal frequency of neural activities are consistent with the instantaneous conditional independence relationships between brain regions in fMRI recordings. Then we propose an enhanced constraint-based method for robust discovery of the underlying causal skeletons, where we include time or condition as an additional variable in the system; it helps avoid spurious causal connections between brain regions and discover time- or condition-specific regions. It also has additional benefit in causal direction determination. Experiments on both simulated fMRI data and real data give encouraging results.

Motivation & Objective

• Address the challenge of weak and unreliable time-delayed causal relations in fMRI due to low sampling frequency.
• Overcome the difficulty in modeling BOLD signal dynamics by linking time-delayed neural activity dependencies to instantaneous fMRI-level conditional independence.
• Account for time-varying and condition-dependent causal structures in fMRI by introducing time or experimental condition as an explicit variable in the causal model.
• Improve causal skeleton recovery and direction identification by leveraging the consistency between neural-level time-delayed and fMRI-level instantaneous conditional independence.
• Develop a principled, robust framework for detecting time- or condition-specific causal relationships in fMRI data

Proposed method

• Propose a simplified fMRI generating process that links time-delayed dependencies in neural activity to instantaneous conditional independence in fMRI recordings.
• Use time or experimental condition as an additional variable in the constraint-based causal discovery framework to model time- and condition-specific dependencies.
• Apply enhanced constraint-based methods to discover causal skeletons while avoiding spurious connections caused by unmeasured time-varying states.
• Leverage the consistency between time-delayed conditional independence at neural activity frequency and instantaneous conditional independence in fMRI to improve causal structure estimation.
• Integrate time or condition as a confounder-like variable to disentangle dynamic causal influences from spurious correlations.
• Utilize the additional variable to enhance causal direction determination by reducing ambiguity in conditional independence tests

Experimental results

Research questions

• RQ1 Can time-delayed conditional independence at neural activity frequencies be reliably mapped to instantaneous conditional independence in fMRI data?
• RQ2 How can time- or condition-specific causal structures be discovered robustly in fMRI when sampling frequency is low and BOLD dynamics are complex?
• RQ3 To what extent does including time or experimental condition as a variable improve the accuracy of causal skeleton recovery in fMRI?
• RQ4 Can the proposed method reduce spurious causal connections caused by unmeasured time-varying states such as attention?
• RQ5 Does incorporating time or condition as a variable enhance causal direction identification in fMRI data?

Key findings

• The time-delayed conditional independence relationships at the neural activity frequency are consistent with instantaneous conditional independence in fMRI recordings, validating the methodological foundation.
• Including time or experimental condition as an additional variable significantly reduces spurious causal connections in fMRI data.
• The proposed method successfully recovers time- or condition-specific causal skeletons, improving robustness to unmeasured time-varying states.
• The method enhances causal direction determination by leveraging the additional variable to resolve ambiguities in conditional independence testing.
• Experiments on simulated fMRI data demonstrate improved accuracy in causal structure recovery compared to standard constraint-based methods.
• Real fMRI data experiments show encouraging results, with the method identifying plausible, condition-specific causal networks