Abstracts

Rationale: To assess the clinical utilization of two independent component analysis (ICA) based artifact reduction methods (AR1 and AR2) in seizure onset determination and assess the validity of the determinations in the context of ictal behavior, other EEG findings, and neuro-radiological findings. Methods: A single-blinded investigation used 23 EEG recordings of seizures from 8 patients. Each recording was uninterpretable with conventional digital filtering because of muscle artifact and were processed using AR1 and AR2 and then reviewed by 26 EEG specialists. EEG readers assessed lateralization, spread, and region of onset and specified confidence for each determination. The two methods were compared on the basis of the number of readers able to render assignments, the confidence measures of the assignments, and the inter-reader agreement of the assignments. Results: Among the 23 seizures uninterpretable with conventional digital filtering, two-thirds of the readers were able to identify the time of seizure onset in 10 ictal EEG recordings using AR1, and in 15 EEGs using AR2. There was a statistically significant increase in the number of readers able to lateralize seizure onset and identify the focus using AR2 as compared with AR1 (p < 0.05), and the confidence measure of this determination increased (p < 0.05). The interclass correlation coefficient for identifying seizure onset was 0.15 (0.11, 0.18) for AR1, and 0.26 (0.21, 0.3) for AR2. The interclass correlation coefficient for lateralizing seizure onset was 0.33 (0.3,0.37) for AR1 and 0.28 (0.25, 0.31) for AR2 indicating that inter-reader agreement for lateralizing seizures following artifact reduction was approximately half that of unobscured seizures evaluated in prior studies. Concordance between the designated laterality of seizure onset using AR2 and behavioral, neurophysiological, and neuro-radiological findings was between 95.9% (85.7, 98.9) for left-sided assignments, and 61.8% (14.9, 68.6) for right-sided assignments. Conclusions: This study demonstrates that ICA-based artifact reduction software methods often lead to interpretations congruent with other clinical data. However, inter-reader agreement and reader confidence for seizure lateralization and localization was marginal, and the differences in interpretation using AR1 and AR2 highlight the need for caution. Utilization of AR2 may improve the validity of ictal EEG artifact reduction. Funding: Dr. Weiss is supported by an Epilepsy Foundation Award Research and Training Fellowship for Clinicians