A template-based method for the estimation of Event Related Potentials using the Bayesian linear model

Abstract

In this work a method for the estimation of Event Related Potentials (ERPs) using the linear model is presented. The method consists of two stages. In the first stage, a template is constructed using the averaged ERP. From this template the design matrix of the linear model is extracted. The second stage is related to the estimation of the coefficients of the linear model. In this stage the bayesian approach is used. However, in our problem the posterior distribution is not easily evaluated and there is need to resort in approximation techniques. One such approach is the Variational Bayesian Methodology. In our study, two prior distributions are used to estimate the ERP. This results in two estimation algorithms having different properties for the coefficients of the linear model. The proposed method is tested in simulated and real ERP data.