Abstracts

Rationale: Magnetoencephalography (MEG) is a tool used in the pre-surgical evaluation of patients with medically intractable focal epilepsy. Recorded interictal MEG spikes can be localized which helps to define the irritative zone and guides additional intracranial EEG recording. This inverse modeling of MEG spikes can be done with a variety of source algorithms but it is unclear which methods are most accurate and correlate with long term seizure freedom. This goal of this study was to compare dipole modeling algorithms (equivalent current dipole [ECD]), current density algorithms (minimum norm estimate [MNE], standardized low resolution electromagnetic tomography [sLORETA], sLORETA-weighted accurate minimum norm [SWARM]), and dipole scanning (multiple signal classification [MUSIC], synthetic aperture magnetometry [SAM], SAM with excess kurtosis [SAM(g2)], and SAM with symmetrically placed virtual sensors [SAM(sym)]) with electrocorticography (ECoG), the resection location, and seizure outcome. Methods: MEG recordings were conducted with a 275 channel CTF magnetometer. Inclusion criteria consisted of (1) pre-surgical MEG between June, 2008 and October, 2011, (2) interictal MEG recording, (3) intracranial ECoG recording completed, (4) resective epilepsy surgery, (5) at least 6 months of post-surgical follow-up. Source localization maps were analyzed by 3 independent reviewers to determine the location of activity for each algorithm according to 5 parameters (left/right, lobe, mesial/lateral, inferior/superior/middle, and anterior/posterior/middle). The same location parameters were used to determine the seizure onset location from the ECoG and the resection location. Seizure outcome was classified according to ILAE criteria (class 1-6) at 6, 12, and 24 months. Results: 230 pediatric patients underwent a pre-surgical MEG from June, 2008 to October, 2011 and 32 patients met the inclusion criteria. Inter-rater reliability for the location of MEG sources using a one sided t-test demonstrated agreement on at least 3/5 location parameters (right/left, lobe, mesial/lateral) for all algorithms (p<0.05). Locations of MEG sources were compared to ECoG; SAM(g2), MUSIC, and ECD were most accurate for identifying the seizure onset location (match of 3/5 location parameters) (23.5%, 20.6%, 20.6% respectively). Locations of MEG sources were also compared to the resection area; sLORETA, MNE, MUSIC, SAM(g2), and ECD were most accurate for identifying the location of the resection (matched on at least 3/5 location parameters) (39.2%, 37.3%, 37.3%, 35.3%, 34.3% respectively). Conclusions: Many MEG source localization algorithms had good general agreement when compared to the site of resection for epilepsy surgery. Dipole modeling (ECD) and dipole scanning (MUSIC, SAM(g2)) algorithms were most accurate for identifying the seizure onset area. The use of multiple inverse modeling algorithms would help to further increase the accuracy of pre-surgical MEG data to guide the placement of subdural grid electrodes.