Meanwhile however, in the previous decade, tens of thousands of hours of expert-scored PSGs have become publicly available ( Zhang et al., 2018, ), and an explosion of algorithmic innovation and increased computing power has opened the possibility to train machine learning and deep learning models on these large datasets. Nevertheless, there have been many attempts, dating back to at least the 1990s, though these generally have not demonstrated robust generalizability. The same indeterminacy also makes algorithmic sleep staging with traditional, analytical approaches difficult. There is self-evidently wide interpretative discretion among sleep scorers as to how to identify these phenomena and how exactly to balance these rules against each other. On top of this, there are extensive contextual rules that specify under what circumstances one stage can follow another, adding a further layer of complexity and subjectivity. Sleep scorers make decisions on the basis of e.g., occipital alpha for a certain period of time, sleep spindles, K-complexes, certain types of eye movements, or overall EEG amplitude, to name just a few. This variability is an inevitable consequence of the relative indeterminacy involved in applying sleep stage criteria to highly complex and variable human polysomnographic data. In addition to the costs of manual classification, there is also substantial disagreement between expert sleep stage scorers (70–80% agreement) and even within the same expert scorer at different times (90% agreement) ( Rosenberg and Van Hout, 2013 Younes et al., 2016 Muto et al., 2018), which introduces a non-trivial degree of variability to both research findings and clinical diagnosis based on manual sleep staging. Established guidelines ( Rechtschaffen and Kales, 1973 Silber et al., 2007) allow for manual classification of PSG data into discrete sleep stages in 30 s increments, but this is a highly laborious process, requiring as much as 2 h to classify a single night’s sleep, even for a trained expert ( Vallat and Walker, 2021). These stages have distinct signatures that can be measured with polysomnography (PSG), which includes the measure of neurophysiology with electroencephalogram (EEG), as well as ocular (EOG) and muscular (EMG) activity. Sleep consists of a rich diversity of neural and physiological stages, which typically unfold in semi-regular cycles throughout the sleep period. These innovations were achieved while simultaneously reaching a level of accuracy equal to, or exceeding, recent state of the art classifiers and human experts, making the GSSC an excellent choice for researchers in need of reliable, automatic sleep staging.Īnalysis of sleep stages for diagnosis of various sleep disorders, as well as analysis on more sophisticated microstructure of sleep like slow oscillations, spindles, and their coupling for research purposes ( Rasch and Born, 2013 for review), has become an important goal in clinical and research context. The GSSC can also be readily integrated into brain-computer interfaces for real-time inference. In addition, the GSSC has been trained to perform well on a large variety of electrode set-ups, allowing high performance sleep staging with portable systems. We introduce the Greifswald Sleep Stage Classifier (GSSC), which is free, open source, and can be relatively easily installed and used on any moderately powered computer. Most of these however have non-trivial barriers to use. With the recent introduction of large, publicly available hand-scored polysomnographic data, and concomitant advances in machine learning methods to solve complex classification problems with supervised learning, the problem has received new attention, and a number of new classifiers that provide excellent accuracy. Manual sleep scoring for research purposes and for the diagnosis of sleep disorders is labor-intensive and often varies significantly between scorers, which has motivated many attempts to design automatic sleep stage classifiers.
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |