Elsevier

NeuroImage

Volume 49, Issue 2, 15 January 2010, Pages 1545-1558
NeuroImage

Six problems for causal inference from fMRI

https://doi.org/10.1016/j.neuroimage.2009.08.065Get rights and content

Abstract

Neuroimaging (e.g. fMRI) data are increasingly used to attempt to identify not only brain regions of interest (ROIs) that are especially active during perception, cognition, and action, but also the qualitative causal relations among activity in these regions (known as effective connectivity; Friston, 1994). Previous investigations and anatomical and physiological knowledge may somewhat constrain the possible hypotheses, but there often remains a vast space of possible causal structures. To find actual effective connectivity relations, search methods must accommodate indirect measurements of nonlinear time series dependencies, feedback, multiple subjects possibly varying in identified regions of interest, and unknown possible location-dependent variations in BOLD response delays. We describe combinations of procedures that under these conditions find feed-forward sub-structure characteristic of a group of subjects. The method is illustrated with an empirical data set and confirmed with simulations of time series of non-linear, randomly generated, effective connectivities, with feedback, subject to random differences of BOLD delays, with regions of interest missing at random for some subjects, measured with noise approximating the signal to noise ratio of the empirical data.

Section snippets

Six problems for causal inference from fMRI

Functional Magnetic Resonance data are increasingly used to attempt to identify not only brain regions of interest (ROIs) that are especially active during perception, cognition, and action, but also the causal relations among activity in these regions (known as “effective connectivities”; Friston, 1994). Modeling of this kind is typically done either by positing a parameterized causal structure a priori and estimating the parameters from data, or by statistical comparison of a few alternative

Addressing the problems using graphical causal models

We propose that modifications of machine learning techniques for graphical causal modeling available for more than a decade (Meek, 1997) provide a basis for addressing the problems we have sketched above. Appendix A presents a more detailed description of graphical causal models and more detail on the algorithms we use to learn these models from data. The software we used is available as freeware in the TETRAD IV suite of algorithms with a graphical user interface at //www.phil.cmu.edu/projects/tetrad

Simulation studies

In view of the theoretical difficulties with inference to latent, non-linear relations between sets of neurons, some investigation by simulation of the accuracy of the IMaGES procedures is needed. To address this, we simulated causal processes in which a large number of unrecorded latent variables, L (e.g., synaptic activity in each of a set of cortical columns), influence one another non-linearly, with feedback, and combine with a hemodynamic response function to produce a multivariate time

Discussion

The IMaGES algorithm is based on a procedure, GES, that has been proven consistent only for feed-forward, acyclic causal structures, whereas the simulated structures have feedback. Despite this, because of the non-linearity of the systems, IMaGES is able to extract the feed-forward structure with reasonable accuracy on large (for fMRI ROI studies) numbers of variables with multiple subject data, missing variables, and varying time delays. The general behavior of the IMaGES algorithm on the

Acknowledgments

This research was supported by a grant from the James S. McDonnell Foundation. We thank Linda Palmer, Peter Spirtes and Richard Scheines for valuable discussions.

References (41)

  • BreakspearM. et al.

    Modulation of excitatory synaptic coupling facilitates synchronization and complex dynamics in a biophysical model of neuronal dynamics

    Netw. Comput. Neural Syst.

    (2003)
  • ChickeringM.

    Optimal structure identification with greedy search

    Mach. Learn. Res.

    (2002)
  • DemiralpS. et al.

    Search for the causal structure of a vector auto-regression

    Oxf. Bull. Econ. Stat.

    (2003)
  • EichlerM.

    A graphical approach for evaluating effective connectivity in neural systems

    Philos. Trans. R. Soc., B

    (2005)
  • FristonK.J.

    Functional and effective connectivity in neuroimaging: a synthesis

    Hum. Brain Mapp.

    (1994)
  • GlymourC.

    The Mind's Arrows: Bayes Nets and Graphical Causal Models in Psychology

    (2003)
  • GlymourC. et al.

    Discovering Causal Structure

    (1987)
  • GrangerC.W.J.

    Investigating causal relation by econometric and cross-sectional method

    Econometrica

    (1969)
  • HansonS. et al.

    The distribution of BOLD susceptibility effects in the brain is non-Gaussian

    NeuroReport

    (2000)
  • HansonS. et al.

    Bottom-up and top-down brain functional connectivity underlying comprehension of every day visual action

    Brain Struct. Funct.

    (2007)
  • Cited by (275)

    • First few seconds for flow: A comprehensive proposal of the neurobiology and neurodynamics of state onset

      2022, Neuroscience and Biobehavioral Reviews
      Citation Excerpt :

      All of these issues present important challenges for neuroscientific investigations of flow. This is especially true when we consider that most neuroscientific methods for linking neural response with psychology or phenomenology are correlational in nature (Ramsey et al., 2010). When considered in conjunction with concerns about reverse inference (Poldrack, 2006), it seems unlikely that neuroscientific investigation will offer clear solutions to these problems.

    View all citing articles on Scopus
    View full text