A Master of Science thesis in Electrical Engineering by Tehereh Zarghami entitled, "Phased Array Technique for Brain Source Localization," submitted in January 2012. Available are both soft and hard copies of the thesis.
Magnetoencephalography (MEG) and Electroencephalography (EEG) are two popular non-invasive techniques that directly measure the magnetic/electric signals generated from the brain's electrical activity, unlike functional measures such as fMRI, SPECT, and PET which reflect blood flow or brain metabolism. Using appropriate volume conductor models for the head, one can theoretically obtain the electric/magnetic fields that would appear outside the scalp as a result of brain neural activity (modeled as current dipoles). This derivation of electric/magnetic fields is known as the forward problem. Finding a proper solution to the forward problem is the first step covered in this thesis. The inverse problem on the other hand is locating the neural activity of the brain which has produced a specific set of measured electric potentials/magnetic fields. The inverse problem is widely known as an ill-posed one, with no unique solution. Moreover, there are major challenges in estimating cerebral sources from EEG/MEG recordings. The effect of physical parameters such as the thickness of skull, conductivity anisotropy, and inhomogeneity of the head affect EEG signals extensively. MEG readings, though less affected by head geometry, are extremely weak and need expensive superconductive sensors and shielded rooms to be acquired. This thesis shows the procedure used for obtaining appropriate forward solutions from the available and dedicated software, using various spherical and realistic head models. Further, to solve the inverse problem, a direction of arrival (DOA) technique recently used in RADAR field is adapted and applied to solving brain source localization problem. This is accomplished through the use of the ReIterative Super-Resolution (RISR) algorithm which is based on a recursive implementation of minimum meansquare error (MMSE) estimation. It is demonstrated that this algorithm can determine the model order, respective DOAs of sources, and their amplitudes from very few samples of data. Results show superior spatial resolution and robustness to correlation between sources. These are important features that are examined through extensive simulations using MEG and EEG data obtained from different head models developed on two different head anatomies. Moreover, the effect of skull thickness and conductivity variations on the localization is investigated for EEG signals. Finally, two calibration techniques have been applied to this algorithm, which improve the localization results in the presence of two types of calibration errors.