Detection of Sleep Oxygen Desaturations from Electroencephalogram Signals

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Sleep apnea is a serious condition often diagnosed using polysomnography.
• Identifying sleep apnea biomarkers from electroencephalogram (EEG) signals could simplify diagnosis.
• Oxygen desaturation events are key indicators of sleep apnea severity.

Purpose of the Study:

• To develop machine learning techniques for identifying sleep apnea biomarkers from EEG signals.
• To detect EEG signals associated with oxygen desaturation during sleep in pediatric patients.
• To explore the potential for a brain-based biomarker for sleep apnea diagnosis.

Main Methods:

• Utilized machine learning algorithms on a large dataset of EEG signals.
• Classified EEG signals as occurring during or not during oxygen desaturation events.
• Investigated the identification of subjects experiencing oxygen desaturations from non-event EEG data.

Main Results:

• Machine learning models achieved an average 66.8% balanced accuracy in classifying EEG signals during oxygen desaturations.
• Demonstrated the ability of machine learning to identify subjects with oxygen desaturations using EEG data.
• Identified potential latent EEG signals indicative of desaturation events.

Conclusions:

• EEG signals contain potential biomarkers for oxygen desaturation during sleep in pediatric sleep apnea.
• Machine learning shows promise in developing a non-invasive, brain-based biomarker for sleep apnea.
• Further research can refine these techniques for improved sleep apnea diagnosis.