Classification of EEG abnormalities in partial epilepsy with simultaneous EEG-fMRI recordings
C Pedreira,A E Vaudano,R C Thornton,U J Chaudhary,S Vulliemoz,H Laufs,R Rodionov,D W Carmichael,S D Lhatoo,M Guye,R Quian Quiroga,L Lemieux
DOI: https://doi.org/10.1016/j.neuroimage.2014.05.009
IF: 5.7
2014-10-01
NeuroImage
Abstract:Scalp EEG recordings and the classification of interictal epileptiform discharges (IED) in patients with epilepsy provide valuable information about the epileptogenic network, particularly by defining the boundaries of the "irritative zone" (IZ), and hence are helpful during pre-surgical evaluation of patients with severe refractory epilepsies. The current detection and classification of epileptiform signals essentially rely on expert observers. This is a very time-consuming procedure, which also leads to inter-observer variability. Here, we propose a novel approach to automatically classify epileptic activity and show how this method provides critical and reliable information related to the IZ localization beyond the one provided by previous approaches. We applied Wave_clus, an automatic spike sorting algorithm, for the classification of IED visually identified from pre-surgical simultaneous Electroencephalogram-functional Magnetic Resonance Imagining (EEG-fMRI) recordings in 8 patients affected by refractory partial epilepsy candidate for surgery. For each patient, two fMRI analyses were performed: one based on the visual classification and one based on the algorithmic sorting. This novel approach successfully identified a total of 29 IED classes (compared to 26 for visual identification). The general concordance between methods was good, providing a full match of EEG patterns in 2 cases, additional EEG information in 2 other cases and, in general, covering EEG patterns of the same areas as expert classification in 7 of the 8 cases. Most notably, evaluation of the method with EEG-fMRI data analysis showed hemodynamic maps related to the majority of IED classes representing improved performance than the visual IED classification-based analysis (72% versus 50%). Furthermore, the IED-related BOLD changes revealed by using the algorithm were localized within the presumed IZ for a larger number of IED classes (9) in a greater number of patients than the expert classification (7 and 5, respectively). In contrast, in only one case presented the new algorithm resulted in fewer classes and activation areas. We propose that the use of automated spike sorting algorithms to classify IED provides an efficient tool for mapping IED-related fMRI changes and increases the EEG-fMRI clinical value for the pre-surgical assessment of patients with severe epilepsy.