ID 142 – Prospective multicenter study of an automatic detection method for burst-suppression patterns. Issue 3 (March 2016)

• Record Type: Journal Article
• Authors: Fürbass, F.; Skupch, A.M.; Koren, J.; Kluge, T.
• Is Part Of: Clinical neurophysiology. Volume 127:Issue 3(2016:Mar.)
• Page Start: e84
• Publication Date: 2016-03
• Subjects: Neurophysiology -- Periodicals; Electroencephalography -- Periodicals; Electromyography -- Periodicals; Neurology -- Periodicals; 612.8
• DOI: 10.1016/j.clinph.2015.11.280
• Languages: English
• ISSNs: 1388-2457

Abstract

Abstract : The occurrence of burst-suppression patterns in the EEG of ICU patients is usually associated with poor outcome if induced by pathological conditions. To avoid exhaustive manual evaluation of the EEG, we developed an automatic and parameter-free method to detect burst-suppression patterns. To assess the detection performance of the method, continuous scalp EEG of 64 consecutive patients was recorded at two ICUs sites resulting in 3962 h of EEG in total (min 6 h, max 192 h). A clinical neurophysiologist was asked to manually review the first minute of each recording hour in the EEG and to mark these segments as burst-suppression or normal EEG. The results of our automatic detection method were compared to the manual annotations to define detection performance by means of sensitivity and specificity. During manual review burst suppression patterns were found in 21 patients, segments without burst suppression were found in 63 patients. The average detection sensitivity was 82% (95% confidence interval from 70% to 90%). The average specificity was 80% (95% confidence interval from 73% to 85%). Our fully automatic detection method for burst-suppression patterns showed high sensitivity and specificity on unselected consecutive long term EEG recordings and will support automated EEG evaluation in the clinical setting.