Beyond traditional visual sleep scoring: massive feature extraction and unsupervised clustering of sleep time series

Abstract

The widely used guidelines for sleep staging were developed for the visual inspection of electrophysiological recordings by the human eye. As such, these rules reflect only a limited range of features in these data and are therefore restricted in accurately capturing the physiological changes that occur during sleep. Here we present a novel analysis framework that extensively characterizes sleep dynamics using over 7700 time-series features from the hctsa software package. We used clustering to categorize sleep epochs based on the similarity of their time-series features, without relying on established scoring conventions. The resulting sleep structure overlapped substantially with that defined by visual scoring. However, we also observed considerable discrepancies between our approach and traditional scoring. This divergence principally stemmed from the extensive characterization by hctsa time-series features, which captured distinctive time-series properties within the traditionally defined sleep stages that are overlooked with visual scoring. Lastly, we report time-series features that are highly discriminative of sleep stages. The data-driven approach identifies powerful individual time-series features that both recapitulate existing knowledge about sleep stage identification and highlight novel time-series analysis methods. Our framework lays the groundwork for a data-driven exploration of sleep sub-stages and has significant potential to identify new signatures of sleep disorders and conscious sleep states, such as dreaming.

Automatic Sleep Stage Classification with Optimized Selection of EEG Channels

“…The combination of feature-classifier-epoch size-data set characteristics will significantly impact the classification accuracy. Possible explorations to extend the model capabilities include unsupervised classification approaches similar to the ones presented in [13], [27], [33]. For these, extensive training using heterogeneous data-sets will be necessary.…”

“…This could mean that the labels can be inconsistent and that the scoring standard is not clear enough. The problem with the scoring standard had been addressed in the study of [13] and should be explored further. And a true unsupervised data-driven approach should be further explored.…”

“…1) Preprocessing: For the input to this model the data was first decomposed in sub-bands using the Fourier Decomposition Method (FDM) [26]. The following frequency bands were extracted: [30][31][32][33][34][35][36][37][38][39][40][41][42][43][44][45][46][47][48] Hz, [12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30] Hz, [8][9][10][11][12] Hz, [4][5][6][7][8] Hz, [0,4] Hz, [0.5-2] Hz, [2][3][4][5]…”

“…The annotations of stages are then performed manually by well-trained human experts, a lengthy and tedious task that also generates considerable inter-rater variability [6]. Many different automatic scoring approaches based on machine learning and deep learning algorithms have been proposed over the years with reasonably good scoring accuracies [7]- [13], but as of today automatic scoring is not a widespread practice in the sleep clinics. Besides accuracy, the adoption of automatic sleep scoring depends among other things, on some practical aspects of the implementation such as the associated computational costs when using all the PSG channels, and the simplicity of the recording device [8], [14].…”

Towards blinded classification of loss of consciousness: distinguishing wakefulness from general anesthesia and sleep in flies using a massive library of univariate time series analyses

“…Historically, candidate markers were often found through visually contrasting electrophysiological recordings, such as electroencephalograms, obtained at varying levels of consciousness [15,16]. Though this approach has led to wellknown markers of depth of anesthesia, it is limited by biases towards groups of time-series features for which differences in level of consciousness are visually clear (for a related issue in sleep research, see [17]). While newer markers have moved away from features which are visually clear, a similar problem applies, wherein researchers investigate features selected based on their own expertise of particular facets of time-series structure.…”