Analysis and visualization of sleep stages based on deep neural networks

Patrick Krauss,Claus Metzner,Nidhi Joshi,Holger Schulze,Maximilian Traxdorf,Andreas Maier,Achim Schilling

doi:10.1016/j.nbscr.2021.100064

Abstract

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 simplify clinical routines. Due to novel open-source software libraries for machine learning, in combination with enormous recent progress in hardware development, a paradigm shift in the field of sleep research towards automatic diagnostics might be imminent. 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, facilitating first assessments on sleep health in terms of sleep-apnea and consequently reduced daytime vigilance. In the following study, we further analyze cortical activity during sleep by determining the probabilities of momentary sleep stages, represented as hypnodensity graphs and then computing vectorial cross-correlations of different EEG channels. 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.

Highlights

Sleep stage scoring is a standard procedure and part of every polysomnographic analysis [1, 2]
We first illustrate how to visualize the representations of EEG data gained from different layers of artificial neural networks. We demonstrate that these complex representations cluster better in higher layers of the artificial neural networks, quantified by the generalized discrimination value (GDV, see [15])
In contrast to handcrafted features we make the neural network to find its own features. This higher order features lead to a good separability of the transformed EEG vectors belonging to different sleep stages

Summary

Introduction

Sleep stage scoring is a standard procedure and part of every polysomnographic analysis [1, 2]. Sleep stage scoring based on physiological signals (EEG: electroencephalography, EMG: electromyography, EOG: electroocculugraphy) is performed by experienced clinicians, which do the classification by hand according to the AASM guidelines [3] This procedure is time consuming and highly prone to errors emphasized by high inter-rater variability [4,5,6]. One core-problem of modern Machine Learning research has met clinical routines, the so called black-box problem [7] or depending on the scientific field the opacity debate [8] This problem can be tackled by using hand-crafted features such as time tags of K complexes or sleep spindles as neural network input [9, 10], (for an example see [11]). A famous example for such phenomena is t-distributed stochastic neighbor embedding (t-SNE, [12]), which is highly unstable and extremely dependent on the initialization and choice of parameters [13]

Methods

Results

Conclusion