Multiple sparse priors technique with optimized patches for brain source localization

Abstract

Localizing brain neural activity using electroencephalography (EEG) neuroimaging technique is getting increasing response from neuroscience researchers and medical community. It is due to the fact that brain source localization has a variety of applications for diagnoses of various brain disorders. This problem is ill-posed in nature because an infinite number of source configurations can produce the same potential at the head surface. Recently, a new technique that is based on Bayesian framework, called the multiple sparse priors (MSP), was proposed as a solution to this problem. The MSP develops the solution for source localization using the current densities associated with dipoles in terms of prior source covariance matrix and sensor covariance matrix, respectively. Then, it uses the maximization of the cost function of the free energy under the assumption of a fixed number of hyperparameters or patches in order to obtain the elements of prior source covariance matrix. This research work aims to further enhance the maximization process of MSP with regard to the free energy by considering a variable number of patches. This will lead to a better estimation of brain sources in terms of localization errors. The performance of the modified MSP with a variable number of patches is compared with the original MSP using simulated and real-time EEG data. The results show a significant improvement in terms of localization errors.