Abstracts

RATIONALE: MEG dipole localization of epileptic spikes is useful in epilepsy surgery to map the extent of abnormal cortex and to focus intracranial electrodes. The most time consuming step is data analysis. Analyzing visually large amounts of data produces fatigue and error, and automated techniques have unacceptable sensitivity vs. specificity characteristics. Therefore sensitive methods for automatic spike analysis are needed. We describe the use of ICA, a method for finding statistically independent components in multisensor data, for separating interictal spikes from background brain activity. We then determine the foci of the subset of these spikes that can be well localized as current dipoles.
METHODS: 9 minutes of spontaneous brain activity were collected using a 68-sensor whole cortex neuromagnetometer (CTF Systems) from an adult female with right temporal lobe epilepsy. Data were sampled at 250 samples/second using 1.25 Hz Butterworth highpass and 70 Hz Butterworth lowpass filters prior to digitization. The data were segmented into 270 contiguous records, each of length 2 seconds. Data were analyzed using the JADE-form of the ICA algorithm. ICA was applied to each 2-second record, and a computer algorithm was used to select the components that exhibited spike-like characteristics. We then applied the MUSIC dipolar source localization algorithm to each of these topographies accepting only those locations that provided at least 95% fit to the topography. The entire procedure was set up to run automatically using our BrainStorm EEG/MEG analysis program.
RESULTS: The ICA algorithm found approximately 2000 spike-like components among the 270 data records. Of these, 56 presented topographies from which a dipole could be localized with at least 95% fit. The data were registered to a T1 weighted subject MR and the sources overlaid on anatomy. Clusters of dipoles were located in left and right temporal lobe, cingulate gyrus and in the vicinity of hippocampus. Other plausible cortical sources were also found as well as implausible sources exterior to the brain and in deep brain structures, possibly due to theta activity and background noise.
CONCLUSIONS: The procedure described is a promising approach to automated screening of spontaneous EEG or MEG data for interictal events. Dipoles localized using this procedure can be clustered to identify candidate epileptogenic regions and to identify data segments for subsequent detailed spatio-temporal analysis. Further refinements of the ICA and component selection algorithms should lead to improvements in the reliability of both spike detection and estimated source locations.
Support: National Institutes of Health grants NS20806, RR13276, MH53213, National Foundation for Functional Brain Imaging, Los Alamos National Laboratory, Huntington Medical Research Institutes, Huntington Hospital, Epilepsy and Brain Mapping Program, and Robert S. and Denise Zeilstra.