Using long-wear electroencephalography to ascertain the variability of Lempel-Ziv Complexity (LZc) measures of consciousness

Abstract

It has been recently claimed that measures of spontaneous electroencephalography (EEG) signal complexity, such as Lempel-Ziv Complexity (LZc), can provide an index of an individual’s level of consciousness. Research and clinical practice are currently limited to unreliable behavioural and physiological measures to indicate consciousness. Therefore, there is significant urgency for an objective, reliable, brain-based measure of consciousness. EEG complexity measures utilise algorithms from Information Theory to quantify the diversity in spontaneous EEG data. These are being used to measure the diverse neural activity which necessarily underlies conscious experience. LZc assesses the complexity of multi-channel EEG data using a compression algorithm. Studies of LZc typically involve comparing conditions of altered consciousness with periods of conscious wakefulness. These studies suggest that the change in complexity observed is reflective of the change in level of consciousness. However, very little is known about how LZc varies, either with or without a corresponding change in consciousness. The present study utilised portable long-wear EEG to record multi-day, continuous EEG data from two participants (a total of 8 days for Participant 1 and 4 days for Participant 2). Data from each participant was analysed independently. A LZc algorithm was used to compute a complexity value for every non-overlapping 10-second segment. Results demonstrated that, as with previous research, LZc during Wake (14-hours during the day, multiple days per participant) is, on average, higher than during sleep (Stage N1, Stage N2, Slow-Wave-Sleep, and REM sleep). However, there is considerable variation surrounding these means. Visualising LZc across Wake revealed a consistent but wide spread of variability around the mean, with a scattering of low LZc values reflected by a negative skew in the data. This also results in a wide range of possible mean LZc values made available from taking samples (between 1 and 120 minutes in duration) during this period. Although this variability reduces with larger samples sizes, even day-to-day, LZc can significantly differ within a person. Regardless of the source of this variability, its presence causes concern due to the overarching clinical motivations and potential practical applications of this measure. These results suggest that LZc may not be indicative of level of consciousness, as previously claimed. The issues raised and addressed in this study are not unique to LZc, but will apply to all complexity algorithms, current and future. With this study, we have shown that long-duration EEG is a successful framework for identifying variability in a complexity measure of consciousness. This information-rich dataset is uniquely capable of exposing and investigating complexity measures, with the additional insight of observing and analysing complexity across time. This study endeavours to redirect discussions of this field and promote the use of this framework to both acknowledge and empirically address all surrounding issues and assumptions. All complexity measures should undergo reliability testing as both a proof of concept and a proof of practice before being utilised in research or clinical applications.