MULTIPLE SOURCE LOCALIZATION USING GENETIC ALGORITHMS

We present a new procedure for localizing simultaneously active multip le brain sources that overlap in both space and time on EEG recordings . The source localization technique was based on a spatio-temporal mod el and a genetic algorithm search routine, The method was successfully applied to the localization of two dipole sources from several sets o f simulated potentials with various signal-to-noise ratios (SNR). The different SMI values resembled evoked responses and epileptic spikes a s commonly seen in the laboratory. Results of the simulation studies y ielded 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 we re based on the dipole arrangements and time activities of data obtain ed during electrical stimulation of the median nerve in human subjects . These studies yielded localization accuracy within 0.1 cm. We also s tudied the localization accuracy of the algorithm using a physical mod el incorporating potential measurements of two current dipoles embedde d in a sphere, In this situation the algorithm was successful in local izing the two simultaneously active sources to within 0.07-0.15 cm.