Automatic sleep stage classification based on subcutaneous EEG in patients with epilepsy

Area of Science:

• Neurology
• Sleep Medicine
• Biomedical Engineering

Background:

• The relationship between sleep patterns and seizure occurrence in epilepsy is well-established.
• Current methods like polysomnography (PSG) offer limited-term sleep monitoring, hindering comprehensive epilepsy management.
• Ultralong-term electroencephalography (EEG) using novel subcutaneous devices presents an opportunity for continuous monitoring of both epileptic activity and sleep architecture.

Purpose of the Study:

• To investigate the feasibility of using subcutaneous EEG recordings for automatic sleep architecture assessment in epilepsy patients.
• To evaluate the accuracy of a subcutaneous EEG-based sleep scoring system compared to traditional long-term video scalp EEG (LTV EEG).

Main Methods:

• Simultaneous long-term video scalp EEG (LTV EEG) and subcutaneous EEG recordings were conducted on four adult epilepsy inpatients over 11 nights.
• Sleep stages were independently scored by an expert using American Academy of Sleep Medicine (AASM) rules for both modalities.
• A machine learning classifier was trained using 30 features from subcutaneous EEG, with LTV EEG serving as the ground truth.

Main Results:

• The automatic sleep stage classifier achieved an average Cohen's kappa of [Formula: see text] using patient-specific cross-validation.
• An awake-sleep classifier demonstrated high performance with 94.8% sensitivity and 96.6% specificity.
• The model showed minor discrepancies in total sleep time, sleep efficiency, and wakefulness after sleep onset compared to manual scoring.

Conclusions:

• This proof-of-concept study demonstrates that two-channel subcutaneous EEG can reliably automate sleep scoring in epilepsy patients.
• The accuracy is comparable to current clinical methods, with the advantage of enabling ultralong-term, patient-specific monitoring.
• This technology holds potential for optimizing epilepsy treatment and improving patient sleep quality through continuous data acquisition.