Analysis and Visualization of Sleep Stages based on Deep Neural Networks

Abstract

<jats:title>Abstract</jats:title><jats:p>Automatic sleep stage scoring based on deep neural networks has come into focus of sleep researchers and physicians, as a reliable method able to objectively classify sleep stages, would save human resources and thus would simplify clinical routines. Due to novel open-source software libraries for Machine Learning in combination with enormous progress in hardware development in recent years a paradigm shift in the field of sleep research towards automatic diagnostics could be observed. We argue that modern Machine Learning techniques are not just a tool to perform automatic sleep stage classification but are also a creative approach to find hidden properties of sleep physiology. We have already developed and established algorithms to visualize and cluster EEG data, in a way so that we can already make first assessments on sleep health in terms of sleep-apnea and consequently daytime vigilance. In the following study, we further developed our method by the innovative approach to analyze cortical activity during sleep by computing vectorial cross-correlations of different EEG channels represented by hypnodensity graphs. We can show that this measure serves to estimate the period length of sleep cycles and thus can help to find disturbances due to pathological conditions.</jats:p>

Analysis of continuous neuronal activity evoked by natural speech with computational corpus linguistics methods

“…In addition, state-of-the-art deep learning approaches may be used as a tool for analyzing brain data, e.g. for creating so-called embeddings of the raw data [122]. Moreover, as proposed by Kriegeskorte [123], our neural corpus can serve to test [124] computational models of brain function [125][126][127][128], in particular models based on neural networks [129][130][131] and machine learning architectures [132,133], in order to iteratively increase biological and cognitive fidelity [123].…”

Interpretation and Further Development of the Hypnodensity Representation of Sleep Structure

“…We conclude, that while the non-linear effect of the softmax function might be desirable in a classification problem, it (partially) discards continuous patterns in the data that might be of medical relevance as well. Other work has presented a finding in the same direction [33], by showing that class separability (related to discreteness) in an automatic sleep staging model highly increased after the last softmax activation. In future work, one might therefore consider other normalizing functions in the classifier, that may replace this (non-linear) softmax function [34].…”

“…In this research, we did not research influences of design choices like data window length (which was fixed to 30 s), depth and width of the models, and the number and type of measurement channels used. Regarding the latter, the authors of [33] visually depicted (supervised) hypnodensities predictions on one EEG channel only, and showed that predictions did not drastically differ dependent on the chosen channel. However, visual inspection seemed to reveal that their singlechannel hypnodensities yielded higher entropy than the presented hypnodensities in this work and in [1].…”

Analysis of continuous neuronal activity evoked by natural speech with computational corpus linguistics methods

“…In addition, state-of-the-art deep learning approaches may be used as a tool for analysing brain data, e.g. for creating so-called embeddings of the raw data (Krauss et al, 2020). Moreover, as proposed by Kriegeskorte and Douglas (2018), our neural corpus can serve to test (Schilling et al, 2018) computational models of brain function (Krauss et al, 2017(Krauss et al, , 2016Schilling, Tziridis, et al, 2020), in particular models based on neural networks (Krauss, Prebeck, et al, 2019;Krauss, Schuster, et al, 2019;Krauss, Zankl, et al, 2019) and machine learning architectures (Gerum et al, 2020;Schilling, Gerum, et al, 2020), in order to iteratively increase biological and cognitive fidelity (Kriegeskorte & Douglas, 2018).…”