Abstracts

Rationale: Non-invasively imaging of epileptogenic zone is of great importance for management of partial epilepsy.Electrical source imaging (ESI) techniques using either E/MEG can be used to help localize epilepsy sources. While ESI techniques have proven useful in determining the location of epileptic activity, objectively estimating the spread (extent) of the underlying activity, is still missing. Currently subjective thresholding of the estimated inverse solution is the only way to determine the spread of activity in distributed models. The goal of this study is to develop and evaluate a novel ESI technique which can objectively determine the extent of the underlying epileptogenic zone from EEG/MEG.Methods: We report a novel iteratively reweighted edge sparse source imaging approach.The proposed method is based on the observation that E/MEG detected brain sources and signals tend to be spatially continuous. This mathematically, translates to sparsity in the gradient domain. Our proposed method is developed in a manner to estimate the epileptogenic zone (location and extent) for partial epilepsy. Computer simulations were performed by selecting random dipoles on the cortex and simulating sources with different sizes at every location. After calculating the scalp potentials (using a generic boundary element head model) and adding noise to the simulated measurements (resulting in SNRs of 10 and 20dB ) the inverse problem was solved and the localization error was found and the size of the estimated and simulated sources (extent) were compared.The algorithm was also tested in 3 partial epilepsy patients. The pre-operational EEGs were recorded using 76 electrodes. Patient specific realistic geometry head models were built from MRIs of the patients. The inter-ictal spikes in each patient were used to estimate the epileptogenic zone using the sparse source imaging.The estimated results were compared to clinical findings such as the seizure onset zone (SOZ), determined by intra-cranial recordings, and resection volume obtained from post-operational MRI.Results: The simulation results show that average localization errors are as low as 5 mm for larger sources and 3 mm for smaller sources. When the estimated and simulated extents were plotted against each other in a single plot, it was found that a one to one relation exists between the estimated and true extents; however a scattering and variance is still observed. The average overlap between the simulated and estimated sources is as high as 85% for larger sources and 70% for smaller sources. This means that the estimated solutions match the simulated sources well. The estimated epileptic zones for the 3 patients tested in this study shows an 80% overlap with the resection area and a 60% overlap with the SOZ.Conclusions: A novel source extent estimation method from noninvasive EEG/MEG was developed.Our results in computer simulations and initial patient study suggest the proposed method is able to rationally image source extent corresponding to epileptic activity.This work was supported in part by NIH EB006433, EY023101 and NSF CBET-1450956.