Modulation of Sleep Architecture by Whole-Body Static Magnetic Exposure: A Study Based on EEG-Based Automatic Sleep Staging
Sleep disturbance has been the main issue for an increasing number of individuals with the progression of society. It can lead to decreased memory and learning, gastrointestinal disorders, depression, and exacerbation of chronic conditions
Approximately 30% of adults and 48% of older adults in particular experience chronic insomnia [2]. Chronic insomnia is difficult to cure using the currently available pharmacotherapy
Therefore, physical therapies have been used to treat chronic insomnia. Electric, magnetic, and electromagnetic fields have been applied to modulate sleep in a series of clinical and experimental studies. Most of these treatment approaches were non-invasive and less stimulant.
Therefore, the application of such therapies is promising, although the effects were not always consistent, nor a clear mechanism of action has been elaborated.
In most studies, questionnaires and self-reported scales are commonly applied to evaluate sleep quality [8]. Among them, the Pittsburgh Sleep Quality Index (PSQI) [9] and the Self-Rating Scale of Sleep (SRSS) [10] are useful tools for sleep-related psychiatric research and practice. The former measures the overall sleep quality during a period, while the latter assesses short-term sleep quality, e.g., the efficacy of a sleep disorder therapy for each night during an experiment.
Sleep has a complex architecture and includes various physiological changes that occur during the period. A person usually experiences four to six sleep cycles per night, which includes different sleep stages. The American Academy of Sleep Medicine (AASM) has divided the sleep process into five stages: awake (N0), non-rapid eye movement (N1-N3), and rapid eye movement (REM) sleep [11]. N3 is slow-wave sleep, which is the most recuperative sleep period and is often indicative of high-quality sleep [12]. Sleep staging is commonly used as an indicator in diagnosing sleep diseases and related psychiatric disorders. In contrast, self-reported questionnaires may implicitly associate with the overall sleep quality but could be undermined by subjectivity; therefore, it is difficult to discriminate the individual sleep stage by using self-reported questionnaires.
Neurophysiological analyses, e.g., using electroencephalogram (EEG), are used in the research of electromagnetic field exposure effects [13,14] and sleep quality determination [15]. Sophisticated paradigms have been developed to process sleep EEG signal into characteristics that reflect sleep rhythms, neural tension, or neural activity [16]. Sleep staging calculates the dwelling time of sleep in each stage by using time-domain signals from multiple electrodes. It traditionally requires extensive manual intervention to discern the specific EEG features. For example, the dataset for an 8-h consecutive sleep may have a volume of 500 MB if sampled at 1000 Hz. In such a case, the results are prone to human error due to fatigue [17]. Therefore, automatic sleep staging is required, and sleep staging based on the machine learning method is a promising alternative [18].
A great number of scientific literatures about automatic sleep staging detection were presented. The majority of these scientific literatures use single-channel EEG recordings for automatic sleep staging [19] and in most cases classified models are built on extracted features. Features are extracted from linear or nonlinear. For improving the classification accuracy and accelerating the model construction procedure, feature selection has become an important step in data preprocessing [20]. There are many feature selection algorithms, including filtering, encapsulation and embedded ones. Decision tree is a typical embedded feature selection algorithm. A decision tree by Liu et al. is suitable for sleep EEG staging due to that it could achieve feature selection for imbalanced data. Selected features are generally used as input for classic algorithms such a support vector machines (SVM), k-nearest neighbor, decision tree (DT), etc. [21]. SVM shows good generalization performance for high dimensional data due to its convex optimization problem [22].
In this study, changes in sleep architecture by static magnetic field exposure (SMFE) were evaluated. Forty-one subjects were randomly divided into two groups (real SMFE group and sham SMFE group) for participation in the experiment for four consecutive nights. Whole-body SMFE was applied by a magnetostatic mattress. During the experiment, sleep EEG was recorded, while PSQI and SRSS were used to report the individual overnight sleep quality. Twenty temporal, frequency and nonlinear metrics were extracted from the labeled sleep EEG by using the Physionet database. A decision tree (DT) was trained using data from this sleep EEG database to select a set of features for sleep staging. The acquired sleep EEG was then classified by a support vector machine (SVM). The purpose of this study is to explore whether there is an ameliorative effect of SMFE on sleep and to explore the adjuvant treatment of chronic sleep disorde