摘要
Although great sensitivities to initial estimates is an inherent feature of iterative dipole optimization algorithms, the study of better initial estimates has been neglected. For convergence to a correct solution, the initial estimates should be extremely close to the desired solution and be attributed to only a single dipole focus. To alleviate the interference of background and multiple loci, the singular value decomposition (SVD) technique is used initially to extract the dominant component of the EEG spike for dipole localization. By observing the three-dimensional topographic mapping, the initial estimates of the dipole parameter set can be computed from the intersection between the null potential plane and from the peak and valley potentials. This work also compares initial estimations of simulation data, including noisy data, noisy data with SVD process and noise-free data. Experimental results confirm that good initial estimates for the dipole parameters are necessary to ensure rapid convergence to the correct solution. Although great sensitivities to initial estimates is an inherent feature of iterative dipole optimization algorithms, the study of better initial estimates has been neglected. For convergence to a correct solution, the initial estimates should be extremely close to the desired solution and be attributed to only a single dipole focus. To alleviate the interference of background and multiple foci, the singular value decomposition (SVD) technique is used initially to extract the dominant component of the EEG spike for dipole localization. By observing the three-dimensional topographic mapping, the initial estimates of the dipole parameter set can be computed from the intersection between the null potential plane and from the peak and valley potentials. This work also compares initial estimations of simulation data, including noisy data, noisy data with SVD process and noise-free data. Experimental results confirm that good initial estimates for the dipole parameters are necessary to ensure rapid convergence to the correct solution.