MSP based source localization using EEG signals

Abstract

The localization of brain sources due to which neural signals are generated is known as brain source localization. These signals are measured by various neuroimaging techniques such as MRI, EEG, PET and MEG. Nevertheless, when the neuroimaging technique is EEG, then it is specifically termed as EEG source localization. This problem is also referred to as EEG inverse problem. This problem is defined by forward problem and inverse problem. Because of ill-posed nature of EEG inverse problem, there exists uncertainty in the solution. This uncertainty in the solution can be reduced by imparting prior information within a Bayesian framework. Hence, Bayesian technique provides some assumptions related to prior information to quantify the solutions. This involves the information of cortical manifold to construct the set of possible regions where the neural activity occurs. This research work discusses and implements the source reconstruction for real time EEG dataset for Bayesian technique (multiple sparse priors (MSP)), classical LORETA and minimum norm techniques. The results are compared in terms of negative variational free energy, intensity level and computational complexity and it is shown that MSP has highest free energy and intensity level as compared to classical methods.