Multiple source localization using genetic algorithms

We present a new procedure for localizing simultaneously active multiple brain sources that overlap in both space and time on EEG recordings. The source localization technique was based on a spatio-temporal model and a genetic algorithm search routine, The method was successfully applied to the localization of two dipole sources from several sets of simulated potentials with various signal-to-noise ratios (SNR). The different SMI values resembled evoked responses and epileptic spikes as commonly seen in the laboratory. Results of the simulation studies yielded localization accuracy ranging from 0.01 to 0.07 cm with an SNR of 10; from 0.02 to 0.26 cm with an SNR of 5; and from 0.06 to 0.73 cm when the SNR was equal to 2. Additionally, two sets of simulations were based on the dipole arrangements and time activities of data obtained during electrical stimulation of the median nerve in human subjects. These studies yielded localization accuracy within 0.1 cm. We also studied the localization accuracy of the algorithm using a physical model incorporating potential measurements of two current dipoles embedded in a sphere, In this situation the algorithm was successful in localizing the two simultaneously active sources to within 0.07-0.15 cm.