Abstracts

Rationale: In clinical magnetoencephalography, single equivalent current dipoles (SECD) are the most commonly used source models for identifying the location of interictal discharges. During the fitting procedure, there are several parameters which are determined by the operator. In addition to selecting the time epoch to analyze, the operator defines the sensors which are to be included in the fit. We have previously investigated the effect of employing gradiometers vs magnetometers or both on source localization of epileptiform abnormalities. In the current study, we investigated the effect of employing various regions of interest. The expertly selected ROIs typically used in practice introduce subjectivity and variability in interpretation, and limit comparison across laboratories. In this study we compared the results from the expertly selected ROIs with the results from three standard ROIs. We speculated that if the error was clinically insignificant, standard ROIs would make the process easier and faster, and that some aspects would be amenable to computerization. Methods: This retrospective study included all 650 epileptic discharges that previously had been manually detected by an expert magnetoencephalographer from 21 MEG studies which had yielded positive results during their clinical interpretation in 2009-2010. The measurements from the 21 patients were performed inside a single layer magnetically shielded room, using a 306 channel whole-head MEG system (Elekta Neuromag, Finland). Before SECD analysis, the data were post-processed using the temporally extended SSS method (tSSS) method to remove artifact. Spikes were manually identified and localized from gradiometers only (i.e. excluding magnetometers) in the course of clinical care while employing the traditional expertly-drawn ROI method. The same timepoints were then reanalyzed employing three standard ROIs. The whole-head ROI consisted of all 204 gradiometers; single hemisphere ROIs consisted of 102 gradiometers. The quadrant ROIs included 78 gradiometers, i.e. more than one-quarter in order to provide some overlap and sufficient degrees of freedom. The results from the standard ROIs were then compared to the SECD results using expert ROIs (original clinical results), in terms of the Euclidean distance from the expert ROI result. Results: The mean error between the expert ROI and the standard whole head, half head, quadrant selections were 11.2, 8.3, and 9.2, respectively. In 15 out of 21 patients, the error employing the standard quadrant ROI was less than 10 mm. Simply selecting the correct hemisphere yielded a better estimate than the quadrant in 13 out of 21 patients. Conclusions: Based on our usual dipole acceptance criteria, a maximum error of 10-12 mm is considered clinically acceptable. Compared to the expert ROI, the error plus two standard deviations of the standard ROI selection was within this range in the majority of patients, supporting magnetoencephalographers impressions that selection of a precise ROI is ordinarily not important.