An expectation-maximization algorithm based Kalman smoother approach for single-trial estimation of event-related...

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Single-trial event-related potential (ERP) estimation is crucial for understanding neural responses.
• Traditional Kalman Filter (KF) approaches rely on pre-specified, manually tuned model parameters, leading to suboptimal and time-consuming estimations.
• Existing methods often overlook the importance of model parameter estimation in ERP analysis.

Purpose of the Study:

• To develop an automated and more accurate method for single-trial ERP estimation.
• To improve upon existing Kalman Filter (KF) based approaches by incorporating Expectation-Maximization (EM) for model parameter estimation.
• To investigate the impact of different noise covariance structures on ERP estimation accuracy.

Main Methods:

• Application of an Expectation-Maximization (EM) based Kalman Smoother (KS) for ERP state and model parameter estimation.
• Integration of maximum likelihood estimation within the EM algorithm to automatically determine model parameters.
• Exploration of various noise covariance structures and selection of the optimal model using Akaike Information Criterion (AIC).

Main Results:

• The proposed EM-based KS approach provides more accurate single-trial ERP estimates compared to methods relying on manual parameter tuning.
• Flexible covariance structures for model noises significantly improve the estimation of inter-trial variability.
• Successful application to chirp-evoked auditory brainstem responses (ABRs) for Wave V detection.

Conclusions:

• The EM-based KS method offers an automated and efficient solution for accurate single-trial ERP estimation.
• Accounting for complex inter-trial variability through flexible covariance structures enhances the reliability of ERP analysis.
• This approach holds promise for improved hearing loss assessment through precise ABR analysis.