Abstracts

Rationale: MEG is increasingly utilized in the presurgical evaluation of patients with intractable epilepsy. While high temporal resolution and lack of distortion by the skull are advantages of MEG, the inverse problem requires all source localization algorithms make a priori assumptions regarding source characteristics, which may affect final predictions of source locations. Although the algorithms perform well in simulation studies with low noise, little is known regarding the relative strengths and limitations of the algorithms for real clinical data. Methods: We recorded MEG interictal discharges for 180 patients (age range 1 mo -26 years) with medically intractable epilepsy. We analyzed all recordings with single or multiple equivalent current dipole models (ECD) and a beamformer, synthetic aperture magnetometry (SAM). For SAM we used an excess kurtosis statistic SAM(g2) to detect regions of increased spikiness and examined the reconstructed time series (virtual sensor waveforms) at peaks in those regions (CTF/VSM DataEditor). More recently we additionally analyzed the recordings in 27 of the patients with dipole scan and distributed dipole models (DDS) (multiple signal classification -MUSIC, minimum norm estimate - MNE, standardized low resolution brain electromagnetic tomography - sLORETA) implemented in CURRY6. We compared each to presurgical scalp interictal/ictal EEG findings. We also compared MEG to ECoG localization for 76 patients who had ECD and SAM only and for 9 patients studied with all 5 algorithms. Results: Interictal discharges had to be chosen carefully for ECD and DDS to obtain representative samples with low background noise; localizations were not possible in the presence of severe noise such as VNS artifact. For ECD and DDS, use of segmented cortical mantle improved localization but also introduced potential errors, related to limitations of the segmentation procedure. DDS tended to localize a single contiguous source region in each patient, but sometimes localized multiple source regions. Localizations differed from interictal and ictal scalp EEG if insufficient spike types were examined. SAM(g2) tended to identify multiple locations. SAM(g2) gave reasonable localizations in the presence of VNS artifact (24 patients) but failed to localize spikes (a) when interictal spikes were very frequent (b) at times when sources were adjacent to prior cerebral infarcts (c) and when the original EEG and MEG recordings had no visible spikes. Virtual sensor waveforms did show spikes near prior cerebral infarcts. Recordings had to not have bursts of muscle artifacts. Use of several methodologies limited errors from each; when multiple algorithms indicated a similar region, the localization more often agreed with ictal onset by ECoG. Conclusions: Different MEG source localization algorithms may highlight different aspects of the irritative zone. When available, use of several different MEG source localization methodologies may improve localization of the irritative zone non-invasively prior to placement of intracranial electrodes and prior to epilepsy surgery.