Nonlinear Bayesian estimation of BOLD signal under non-Gaussian noise

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Computational Neuroscience

Background:

• The blood oxygenation level dependent (BOLD) signal is crucial for functional magnetic resonance imaging (fMRI) analysis.
• Hemodynamic models, inspired by fluid dynamics, interpret BOLD signals by integrating physiological changes.
• Standard fMRI data analysis often assumes Gaussian noise, which can limit filter performance in real-world scenarios.

Purpose of the Study:

• To develop and evaluate a novel data assimilation scheme for BOLD signal modeling under non-Gaussian noise conditions.
• To improve the accuracy of state and parameter estimation in hemodynamic models.
• To address the limitations of traditional filters when dealing with non-Gaussian noise in fMRI data.

Main Methods:

• A hemodynamic model comprising nonlinear differential equations (process model) and a weighted sum of physiological variables (measurement model) was utilized.
• A new data assimilation filter, the Multi-Agent Gaussian Sum Filter (MAGSF), was proposed to handle additive non-Gaussian (e-mixture) noise.
• The performance of MAGSF was compared against the Extended Kalman Filter (EKF) using joint optimal Bayesian filtering.

Main Results:

• The proposed MAGSF filter demonstrated superior performance in estimating states and parameters of the hemodynamic model compared to the EKF.
• Analyses using both synthetic and real fMRI data confirmed the enhanced capabilities of MAGSF under non-Gaussian noise.
• The study highlights the impact of noise assumptions on filter efficacy in BOLD signal modeling.

Conclusions:

• The MAGSF offers a more robust approach for BOLD signal modeling and analysis in fMRI, particularly when noise deviates from Gaussian assumptions.
• Accurate hemodynamic modeling under realistic noise conditions is essential for advancing neuroimaging research.
• This work provides a valuable tool for the neuroimaging community to improve the reliability of fMRI data interpretation.