Abstracts

RATIONALE: Source localization algorithms have been applied to the problem of identifying intracranial sources of seizure for many years. There are now multiple methods available but few comparative studies across methodologies. One of the problems with these studies is that the [dsquote]gold[dsquote] standard of performance for validation of algorithms is that seizures are eliminated when the predicted site of the source is removed surgically. This evidence is important support for the accuracy of the source localization techniques, but the differences in resolution between some of the techniques is such that surgical outcome is not a sufficiently refined approach to reveal differences in spatial predictions. Therefore, we have performed a feasibility study using three alternative methods for quantifying differences in prediction between source localization algorithms
METHODS: All three methods depend upon the simultaneous collection of intracranial and extracranial data. The first method is to compare spatial separation between two stimulated electrodes on the intracranial grids during cortical mapping, to the predicted spatial separation based upon surface data. The second is to compare electrode spatial separation of independent, but spatially distant, spontaneous spikes or slow waves on the intracranial grid to that predicted by the localization algorithms. The third method is to compare the absolute position of the electrode site where the seizure was initiated to that predicted by the algorithm. These techniques were applied to the following algorithms: Least squares minimization linearly constrained minimum variance filters, MUSIC and LORETA.
RESULTS: Each of the methods provides data on the performance of the algorithms [ndash] with the exception of LORETA (depending on the condition). The methods are more powerful when applied to situations in which the separation of the electrodes on the intracranial grid is at least twice the limits of resolution of the algorithms. The signal to noise ratio of the targeted events has a major impact on the reproducibility of the results. The most difficult method to apply is the prediction of the absolute position since it depends on the adequacy of the forward model and the accuracy of co-registration.
CONCLUSIONS: The purpose of this report was to demonstrate the feasibility of more refined methods of validation of source localization algorithms. The limits of resolution for the [dsquote]surgical[dsquote] validation are determined by the size of the resection, while the limits of resolution for the proposed methods are limited by inter-electrode distance on the grid. We conclude that validation according to outcome of surgical intervention can be replaced by more precise methods.
Support: Falk Medical Trust Foundation.