Abstract
This thesis is concerned with the processing of MagnetoEncephaloGraphic (MEG) signals before their further analysis for clinical purposes. An overview of recent methods that have been applied to brain signals is first presented. The area of interference elimination is covered then, as the MEG signals suffer from the heart interfering magnetic field which in the majority of the experimental situations outweighs the signal of interest. The framework of a two step algorithm which first identifies the interference and then eliminates it by orthogonal projecting it to the contaminated signal is proposed. Next, the restoration of an MEG-like signal of interest buried in coloured noise is attempted by adopting a multi-model representation of the mixed signals. The noise is modelled an a autoregressive process and the deterministic signal of interest as a Markov Random process assuming piecewise linearity. The Simulated Annealing method is adopted for the restoration procedure. Furthermore, multi-resolution is used in order to accelerate the algorithm. Finally, a solution to the inverse solution is attempted by using the Singular Value Decomposition technique to decompose the measurement signal space to two subspaces, one dominated by the heart and another one dominated by the brain. Throughout the thesis results are presented with both synthetic and real data in order to illustrate the validity and usefulness of the proposed algorithms.