Biomedical Signal Processing for Improved Diagnosis of Sleep Disorders and Brain Disease

Sleep is a complex biological state consisting of a multitude of distinct physiological responses in several different systems of the body. Examinations of sleep are performed by sleep doctors and technicians using polysomnography (PSG), which is an umbrella term covering recordings of various physiological variables, such as electroencephalography (EEG, brain activity), electrooculography (EOG, eye movements), electromyography (EMG, muscle activity), electrocardiography (ECG, heart rate variability), respiratory airflow, pulse oximetry, actigraphy and more.

The information from these recordings is used to determine both the temporal progression of sleep as well as events during sleep that either help determine a specific sleep stage, such as the presence of sleep spindles or K-complexes, or aid in diagnosis of a specific disease or condition, such as sleep apnea events or movements of the lower extremities. Sleep experts assign discrete labels to each 30 s corresponding to either wakefulness (W), rapid eye movement (REM) sleep, or one of three stages of non-rapid eye movement (NREM) sleep, based on a set of rules and guidelines defined by the American Academy of Sleep Medicine (AASM).

However, these examinations are time-consuming and prone to errors caused by subjective interpretations of both the physiological signals and scoring rules. Studies of the agreement between scorers of varying levels of experience scoring the same datasets have shown an average agreement between 63.0% and 85.2% depending on the specific sleep stage . Several attempts have already been made at automating sleep scoring, assisting in diagnosis and detecting specific events during sleep, but none have had any widespread use or impact in research or in a clinical setting. This is partly because most research in this area is based on relatively small and homogeneous populations.

This PhD project aims to bridge the gap between clinical research, sleep medicine and technology by researching methods to develop a data-driven clinical support system capable of fully automatic and comprehensive analysis of sleep patterns. The system will be designed and developed based on a large-scale database containing recordings and medical information from multiple patient groups from Danish and American databases. It is the hope, that this system can assist health care professionals in improving early diagnosis of sleep disorders and brain diseases from a single overnight PSG recording, thereby eliminating or reducing the subjective element in sleep research.

The project is co-funded by Stanford University and DTU.