BOLD Signal Estimation Based on Dual Particle Filter

Functional magnetic resonance imaging (fMRI) methods measure neuronal activity-induced changes indirectly by the blood oxygenation level dependent (BOLD) effect. The most comprehensive model to date of the BOLD signal is formulated as a mixed continuous discrete time system of nonlinear stochastic differential equations. Previous approaches have been based on linear approximations of the dynamics, which are limited in their ability to capture the inherent nonlinearities in the physiological system. A dual particle filter is proposed to simultaneous estimate the hidden physiological states and the system parameters in this paper, the sufficient statistics is adopted to deal with sampling degeneracy phenomena and the beta distribution which makes good use of prior knowledge as well as avoids tail draws for the parameter is used to fit the parametric posteriori probability density function. This approach applied to phantom data and human subjects. The results showed that state estimates via simulation are accurate, robust and efficient in comparison to linearization-based technique and physiologically reasonable parameter estimates are generated for experimental fMRI data.