Stochastic dynamic causal modelling of FMRI data with multiple-model Kalman filters

Area of Science:

• Neuroimaging analysis
• Computational neuroscience
• Biosignal interpretation

Background:

• Dynamic Causal Modelling (DCM) is a framework for studying brain connectivity using neuroimaging data like fMRI.
• Existing DCM methods are being adapted to incorporate stochastic disturbances.
• This work focuses on advanced methods for neural signals and images.

Purpose of the Study:

• To introduce the Multiple-Model Kalman Filtering (MMKF) technique for stochastic identification within the DCM framework.
• To assess MMKF's performance in discriminating between different hypothetical brain connectivity structures.
• To compare MMKF against a similar deterministic identification model.

Main Methods:

• Integration of stochastic DCM equations.
• Development of a MMKF algorithm for model selection based on stochastic DCM.
• Monte Carlo simulations to evaluate MMKF's ability to distinguish connectivity structures and estimate hidden states.

Main Results:

• MMKF successfully identified the correct brain connectivity model structure from plausible alternatives.
• The stochastic MMKF approach demonstrated superior performance over deterministic methods.
• MMKF showed enhanced accuracy in both selecting the correct connectivity and estimating hidden states.

Conclusions:

• The MMKF approach is applicable to studying effective brain connectivity with DCM.
• Stochastic formulations using MMKF are particularly beneficial for brain connectivity analysis.
• This method advances biosignal interpretation for neural signals and images.