Joint state and parameter estimation of the hemodynamic model by particle smoother expectation maximization method

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Signal Processing

Background:

• Accurate estimation of hemodynamic model parameters and states is crucial for fMRI analysis.
• Existing methods for joint state and parameter estimation in hemodynamic models are limited.
• Blood oxygen level dependent (BOLD) signals are fundamental to fMRI data.

Purpose of the Study:

• To implement and evaluate the particle smoother expectation maximization (PSEM) algorithm for robust estimation of hemodynamic model parameters and states.
• To compare the performance of PSEM against state-of-the-art methods in hemodynamic model inversion.

Main Methods:

• Implementation of the particle smoother expectation maximization (PSEM) algorithm, a maximum likelihood-based method.
• Joint state and parameter estimation using fMRI BOLD signals.
• Comparative analysis of PSEM against the square-root cubature Kalman smoother (SCKS) and other established algorithms (DEM, EKF, LL).

Main Results:

• PSEM demonstrated superior accuracy in both state and parameter estimation compared to SCKS.
• PSEM outperformed previous sequential Monte Carlo methods, EKF, and LL filters in state estimation.
• SCKS was found to be more effective than DEM, EKF, LL, and particle filters.

Conclusions:

• The PSEM algorithm represents the most accurate method currently available for system identification and state estimation in hemodynamic model inversion.
• PSEM significantly advances the field of fMRI data analysis by providing more reliable parameter and state estimates.
• Further comparisons with methods like Tikhonov-regularized Newton-CKF (TNF-CKF) are warranted.