Abstracts

Rationale: Localization of interictal discharges from MEG recordings is most commonly done by calculating single equivalent current dipole (SECD) models, usually resulting in multiple dipoles. It is well known that the extent of these dipoles does not reflect the size of the epileptogenic zone, but rather the signal to noise ratio. Although finding a tight cluster is not necessary for accurate source localization, cluster tightness is frequently invoked as a measure of confidence in dipole localization. Despite the ubiquity of this concept, the literature contains no systematic analysis or definition.Methods: We developed a program which imports standard dipole files and produces a statistical and visual representation of the dipoles. Firstly, a k-means iteratively partitioning process groups the dipoles into clusters. The location of the centroid of each cluster is calculated, along with the average distance to each dipole. For display and measurement purposes, a sphere is drawn showing a cluster boundary, the mean distance of all dipoles from the centroid plus 1 standard deviation. For calculation of the orientation cone, all of the dipoles within a given cluster are translated to a common origin so that their average solid angle can be computed. For visualization, a three-dimensional cone oriented along this axis and with a vertex angle based on the average deviation, plus 1 s.d., is generated. These graphical elements, the cluster sphere and the orientation cone for each cluster, can be shown, along with the individual dipoles, coregisteed to the patient s MRI. A cluster boundary of 10 mm and an orientation cone of 10 degrees were empirically chosen as the maxima for classification of a tight cluster and consistent orientation respectively. In a pilot application, we studied 50 patients who had positive MEG findings and subsequent ICEEG between 2008 and 2010. MEG/ICEEG anatomical agreement and MEG cluster resection were correlated to epilepsy surgery outcome.Results: 48 clusters were identified, 15 of which were tight clusters and 20 had consistent orientation. There was agreement between the MEG and ICEEG localization in 75% when the clusters were tight. In 36 patients who underwent resective surgery, 70% were seizure-free when the clusters were tight or the orientations were consistent. However, some patients outside the 10 mm / 10 degree cluster bounds also did well.Conclusions: Implementation of an automated tool for statistical calculation of cluster boundary and orientation cone has allowed us to quantify 2 important characteristics of dipoles identified during SECD analysis of MEG data. Our initial application of this method has shown it useful for objectively describing MEG results and for comparing MEG localization with ICEEG and surgical resection. More refined extensions of this approach should take into account parameters of the dipole fits. Prospective evaluation of these new methods and epilepsy surgery outcome will allow us to evolve the current empirical thresholds into clinical standards.