Large-scale dynamic modeling of task-fMRI signals via subspace system identification

Area of Science:

• Neuroscience
• Systems Biology
• Control Theory

Background:

• Task-based functional magnetic resonance imaging (fMRI) generates complex time series data.
• Understanding the dynamical input-output relationships in the brain is crucial for advancing neuroscience.
• Existing methods may not fully capture the intricate dynamics of brain activity during tasks.

Purpose of the Study:

• To develop accurate, large-scale dynamical models from task-based fMRI time series.
• To characterize the dynamic behavior of the brain using control-theoretic analysis.
• To enable quantification of brain responses to external task-related inputs.

Main Methods:

• Extended subspace system identification for deterministic and stochastic state-space models with external inputs.
• Employed control-theoretic tools for dynamic behavior characterization.
• Utilized probabilistic inversion via joint state-input maximum likelihood estimation for validation.
• Applied efficient algorithms to high-dimensional optimization problems from dense fMRI data.
• Extended subspace methods for multi-subject analyses.

Main Results:

• Accurate large-scale dynamical models were produced from fMRI time series.
• Quantified input-output transfer functions between task conditions and cortical regions.
• Identified models demonstrated impulse response functions consistent with hemodynamic responses.
• Captured common dynamical characteristics across multiple subjects.
• Successfully estimated task stimuli timing from observed brain outputs.

Conclusions:

• Dynamical input-output models can accurately represent fMRI data.
• These models provide a basis for control-theoretic approaches in neuromodulation and neurofeedback.
• The findings support the use of these models for therapeutic applications in self-regulation.