Abstracts

AUTOMATED ARTIFACT DETECTION IN INTRACRANIAL EEG

Abstract number : 3.115
Submission category : 3. Neurophysiology
Year : 2012
Submission ID : 16481
Source : www.aesnet.org
Presentation date : 11/30/2012 12:00:00 AM
Published date : Sep 6, 2012, 12:16 PM

Authors :
B. Brinkmann, V. Vasoli, J. Echauz, B. Litt, G. Worrell

Rationale: Long-term, continuous, wide-bandwidth recording of human electrophysiology from hundreds of electrode channels shows promise for delineating the epileptogenic zone in focal epilepsy via detection and mapping of high frequency oscillations, microseizures, unit action potentials, and interelectrode synchrony. Data artifacts due to human motion, electromagnetic induction, and poor electrical contact are inevitable in recordings that can last over a week. The large amount of data generated by this approach precludes comprehensive human review for data quality, and regions of artifact must be excluded from automated analysis by electrophysiological detection algorithms. Methods: Data condition number (DCN), independent component analysis (ICA), and line length have been proposed as automatic methods for detecting time segments of poor data quality in EEG recordings. However, no existing method is universally accepted, and some current methods do not differentiate between recorded data segments where signal is uniformly poor versus segments where artifacts are limited to only a few channels, or one particular electrode. We describe and validate a modification to the DCN data quality metric that can be applied to single channels or groups of channels to provide a better understanding of data quality issues. The relative data condition number (rDCN) for a given channel or group of channels is calculated by subtracting the DCN with the channel(s) in question omitted from the DCN calculated on the entire group. Results: The rDCN is validated and demonstrated on data from the NIH International Epilepsy Electrophysiology Database (www.ieeg.org) and is shown to successfully differentiate between data segments with global artifacts (affecting all channels) and artifacts confined to single channels or groups of channels. Its performance is characterized in comparison to the performance of other existing data quality algorithms. Conclusions: Automated detection and analysis of EEG recording artifact is an important component to automated detection and analysis in long-term EEG recordings and ambulatory EEG monitoring systems. The rDCN has the potential to accurately and reliably identify segments and channels with poor data quality without user intervention.
Neurophysiology