Beta 1


Title EEG Source Localization using a Hierarchical Bayesian Approach
Author Stahlhut, Carsten (Intelligent Signal Processing, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Supervisor Hansen, Lars Kai (Intelligent Signal Processing, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Institution Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
Thesis level Master's thesis
Year 2008
Abstract In this thesis five models are proposed for the inverse EEG problem with the use of a hierarchical Bayesian approach. For one of these models an Expectation Maximization (EM) algorithm is derived and for the rest Variational Bayesian (VB) algorithms are derived. The algorithms use Automatic Relevance Determination (ARD) priors in order to automatic find the most active current sources. The main contribution of this thesis compared to previous work, is an incorporation of an uncertainty of the head model that relates the current sources in the brain to the measured scalp potentials, in the VB framework. Furthermore, the derived algorithms allow the possibility to deal with time-dependent variance of the sources and different noise levels at the channels. Extensive simulations on synthetic data with the use of a random generated head model and a realistic one have been performed in a systematic way, in order to reveal the influence of the model parameters on the inverse EEG problem. The results on synthetic data and a random generated head model show that one of the VB algorithms without uncertainty of the head model included performs slightly better than a similar algorithm with the uncertainty included. However, results on synthetic data with a realistic head model indicate that the VB algorithm with the uncertainty of the head model included leads to a better localization of the simulated sources, when compared to the similar VB algorithm without the uncertainty included. Future work would include simulations on real EEG data to show the applications of the algorithms.
Series IMM-M.Sc.-2008-12
Fulltext
Original PDF ep08_12.pdf (4.71 MB)
Admin Creation date: 2008-02-19    Update date: 2010-10-28    Source: dtu    ID: 211023    Original MXD