Whole-brain causal discovery using fMRI

Area of Science:

• Neuroscience
• Computational Biology
• Network Science

Background:

• Discovering causal relationships in functional magnetic resonance imaging (fMRI) data is a significant challenge.
• Existing methods like Granger causality and dynamic causal modeling struggle with contemporaneous effects and latent common causes.
• Causal structure learning methods face scalability issues and often require acyclic assumptions, limiting their application to large-scale brain networks.

Purpose of the Study:

• To address limitations in current fMRI causal discovery methods.
• To develop a scalable and accurate method for inferring causal relationships from large-scale, low-resolution time-series fMRI data, incorporating feedback.
• To establish a new standard for causal discovery in whole-brain fMRI analysis.

Main Methods:

• A taxonomy and comparative analysis of existing fMRI causal discovery methods were performed on simulated data.
• A novel constraint-based method, Causal discovery for Large-scale Low-resolution Time-series with Feedback (CaLLTiF), was developed.
• CaLLTiF utilizes conditional independence tests on contemporaneous and lagged variables to identify causal links.

Main Results:

• CaLLTiF demonstrated superior accuracy and scalability compared to existing methods on simulated fMRI data from the macaque connectome.
• Analysis of human resting-state fMRI revealed highly consistent causal connectomes across individuals using CaLLTiF.
• Learned causal connectomes exhibit a top-down causal flow from attention and default mode networks to sensorimotor networks, with effects dependent on Euclidean distance and dominated by contemporaneous interactions.

Conclusions:

• CaLLTiF represents a significant advancement in causal discovery from whole-brain fMRI data.
• The method overcomes key limitations of previous approaches, offering improved accuracy and scalability.
• This work sets a new benchmark for future research in understanding brain connectivity and function through causal inference.