Abstracts

This is a Late-Breaking abstract.

Rationale: Magnetoencephalography (MEG) is a valuable and noninvasive FDA-approved tool for source localization of the epileptogenic zone during the presurgical workup of drug-resistant epilepsy. Clinical MEG analysis usually relies on equivalent current dipole (ECD) fitting to identify sources of interictal epileptiform discharges (IED). However, this is a time-consuming and laborious analysis that requires highly trained physicians. Other MEG analysis methods, such as the kurtosis beamforming analysis, are more easily automated. It is a spatial filtering method that estimates the kurtosis of each region’s time series in source space, with an apparent increased value in areas containing epileptogenic activity versus the regions with baseline brain activity. There are different open-source pipelines to process kurtosis. This study aimed to compare MEG seizure onset zone (SOZ) source localization using a modified pipeline of the kurtosis beamforming against the gold standard intracranial electroencephalography (iEEG).

Methods: Clinical resting-state MEG scans were performed using a 306-channel MEGIN Triux-Neo. Eight drug-resistant epilepsy patients were selected. Temporal signal space separation (tSSS) was performed on MEG data using MaxFilter software. A neurophysiologist identified IED in the sensor time series, and the ECD fitting was performed in MEGIN. MEG data with their corresponding T1 weighted Magnetic Resonance Imaging (MRI) were exported to MATLAB. Two open-source codes were available for source localization using the kurtosis beamformer method in the Fieldtrip toolbox. These two were combined to make the process faster and more user-friendly with the following steps: (1) Determination of the fiducial point coordinates by an automated co-registration of MEG and MRI. (2) A band-pass filter of 20-70 Hz to the tSSS resting-state file. (3) Selection of an appropriate time segment to ensure that the analysis contained as many spikes as possible while trying to avoid artifacts. (4) Construction of a source model using a 5 mm resolution. (5) Computation of the beamformer virtual channels and kurtosis using linear constrained minimum variance (LCMV). (6) Finally, the virtual electrode correspondent raw data from the candidate sources were visually inspected to verify the presence of spikes.

Results: Our code takes the anatomical co-registration result and generates the source images for calculating the beamformer output. It localized candidate sources using a user-friendly and semi-automated technique. Kurtosis beamformer candidate sources overlapped or were adjacent to the seizure onset zones identified by iEEG (Figure1). In addition, some patients also yielded strong peaks in areas corresponding to areas of increased epileptogenicity, also considered to be clinically significant (Figure 2).

Conclusions: Utilizing this modified processing method, we demonstrated that the kurtosis source localization agreed with SOZ identified by iEEG. Based on our preliminary results, the kurtosis beamforming provides clinically meaningful results with a more automated process than the ECD method.

Funding: Hoblitzelle Foundation