Integrated approach of an artificial neural network and numerical analysis to multiple equivalent current dipole source localization.

Abstract

The authors have developed a PC-based multichannel electroencephalogram (EEG) measurement and analysis system. This system enables us (1) to simultaneously record a maximum of 64 channels of EEG data, (2) to measure three-dimensional positions of the recording electrodes, (3) to rapidly and precisely localize equivalent current dipoles (ECDs) responsible for the EEG data, and (4) to superimpose the localization results on magnetic resonance images. A new neural network and numerical analysis (NNN) approach to ECD localization is described which integrates a feedforward artificial neural network (ANN) and a numerical optimization (Powell's hybrid) method. It was shown that the ANN method has the advantages of high-speed localization and noise robustness, because in this approach: (1) ECD parameters are immediately initialized from the recorded EEG data by the ANN and (2) ECD parameters are accurately refined by the hybrid method. Our multiple ECD localization method was applied to sensory evoked potentials and event-related potentials using the present system.