Burst Suppression Segmentation of EEG Using Adaptive Binarization in Time and Frequency Domains

This paper proposes a method for burst suppression segmentation using adaptive EEG binarization in both time and frequency domains. Since the dynamic range of EEG amplitude is very wide and can vary from one recording to another owing to the physical conditions of the subjects and measuring factors, such as electrode types, electrode locations, impedance, and amplifiers, conventional quantitative EEG (qEEG) features for segmentation sensitively vary in proportion to the EEG amplitude and result in erroneous segmentation. EEG binarization was applied to solve the problem of a wide dynamic range. Through the restriction of all different dynamic values of EEG signals to 0 and 1, subsequent signal processing techniques are independent of the dynamic range. Additionally, binarization may reduce the computational time of processing data. We employed the ordered statistic constant false alarm rate (OS-CFAR) algorithm for adaptive binarization. For burst suppression segmentation, maximum likelihood estimation was conducted using Gaussian models for feature values of burst and suppression segments. To investigate the segmentation performance of the proposed method, we evaluated the accordance to visual segmentation and quantified the error of burst suppression ratio estimation. Generally, the proposed binarization method was shown to be beneficial for both increased segmentation accordance and decreased error of estimated burst suppression ratio.