A review of automated sleep stage scoring

Sleep stage scoring is an important aspect of sleep medicine and research. The literature for automated sleep stage scoring contains a wide range of methods. A major goal of this study is to give an overview on how these methods are combined such that the automated sleep stage scoring functionality emerges. We discuss and advise on signal processing techniques that are used for automated sleep stage scoring. This discussion starts with a review of benchmark databases which contain measurement data and corresponding expert sleep stage scoring results. The data processing techniques can be grouped into one of four main categories: pre-processing, feature engineering, feature selection, and classification. After the individual signal processing techniques are introduced, we move on to review scientific work on automated sleep stage scoring. At the heart of automated sleep stage scoring are the classification algorithms which provide decision support for human experts. The classification is realized by training and testing artificial intelligence algorithms. During the review, we have learned that properties such as safety, reliability, and functionality of the decision support relate to the choice of artificial intelligence algorithm. Especially choosing between traditional machine learning and deep learning (DL) algorithms will shape both validity and transferability of test results.