Chaotic Simulated Annealing Applications In Eeg Dipole Localization Problem

The EEG inverse problem is the estimation of the neural current sources underlying a measured distribution of scalp potentials. It is a common practice to model neural activity by one or more equivalent current dipoles, where each dipole is completely characterized by three location parameters and three moment parameters. The inverse problem is then defined as the estimation of the parameters of dipoles whose potentials best fit the actual measurements in a least-squares sense, which is a very difficult nonlinear optimization problem strongly depending on the optimization technique. A local optimization method for such a problem is very susceptible to being trapped in a local minimum. Global optimization methods are currently more common due to their ability to converge to the global minimum, yet by which computational costs are much higher. Within the past couple of years, a lot of work has been done to develop more effective global algorithms, most of which have been limited to improvement by means of combination of the local and global methods. In this present paper, to the EEG inverse problem we have extended the efforts to introduce a chaotic optimization algorithm which is widely applied on artificial neural network ,communication network,geophysical inverse problem.In this paper, we present and evaluate a robust and efficient optimization approach, named chaotic simulated annealing(CSA) algorithm for this problem. Its superiority of effectiveness over the traditional nonlinear technique were tested and demonstrated through computer simulations. The following focus of the present paper is on the application of CSA to localizing dipoles accounting for the early event-related potential (ERP)effects elicited by spatial visual attention, aiming to perform more efficient estimation of the neural generators. The results provide support to current neurophysiological hypotheses on spatial attention-related visual signal processing. Compared with the traditional stochastic optimization algorithm, CSA converged more quickly and accurately to the global minima and proved a promise global optimization method of high adaptability and feasibility.