Bayesian Framework based Brain Source Localization Using High SNR EEG Data

The multipurpose application of brain source localization in the domain of biomedical engineering has aggrandized the ways for its further development for various healthcare applications. Various brain regions are activated due to different mental and physical tasks. These sources can be localized using different optimization algorithms. This localization information is usable for diagnoses of brain disorders such as epilepsy, Schizophrenia, depression and Alzheimer. The brain signals are recorded through neuroimaging techniques such as MEG, EEG, fMRI and PET etc. Nevertheless, when EEG signals are used to reconstruct the active brain sources, then its termed as EEG source localization. The localization involves two phases: forward modeling and inverse modeling. The forward modeling is carried out to model the head using various numerical techniques. Some of them are finite element method (FEM), boundary element method (BEM) and finite volume method (FVM). Furthermore, the solution of inverse problem is realized through usage of various optimization techniques such as minimum norm estimation (MNE), low resolution brain electromagnetic tomography (LORETA), multiple signal classification (MUSIC) and multiple sparse priors (MSP). This research work discusses the results based on synthetically generated EEG data at an SNR level of 12 dB with Gaussian noise added linearly in data matrix. For forward modelling, boundary element method with 1 μA dipole amplitude at dipole frequency of 20Hz is utilized. The dipoles’ location is set randomly at 2000 and 5700 at CTF. The techniques mentioned above are applied on data and are compared with latest multiple sparse priors (MSP). The evaluation of results is done based on comparative analysis of free energy and localization error. The results shows superiority of MSP over classical algorithms.