Automatic Sleep Monitoring Using Ear-EEG

Takashi Nakamura, Valentin Goverdovsky, Mary J. Morrell, Danilo P. Mandic

Abstract

The monitoring of sleep patterns without patient's inconvenience or involvement of a medical specialist is a clinical question of significant importance. To this end, we propose an automatic sleep stage monitoring system based on an affordable, unobtrusive, discreet, and long-term wearable in-ear sensor for recording the Electroencephalogram (ear-EEG). The selected features for sleep pattern classification from a single ear-EEG channel include the spectral edge frequency (SEF) and multi-scale fuzzy entropy (MSFE), a structural complexity feature. In this preliminary study, the manually scored hypnograms from simultaneous scalp-EEG and ear-EEG recordings of four subjects are used as labels for two analysis scenarios: 1) classification of ear-EEG hypnogram labels from ear-EEG recordings and 2) prediction of scalp-EEG hypnogram labels from ear-EEG recordings. We consider both 2-class and 4-class sleep scoring, with the achieved accuracies ranging from 78.5 % to 95.2 % for ear-EEG labels predicted from ear-EEG, and 76.8 % to 91.8 % for scalp-EEG labels predicted from ear-EEG. The corresponding kappa coefficients, which range from 0.64 to 0.83 for Scenario 1 and from 0.65 to 0.80 for Scenario 2, indicate a Substantial to Almost Perfect agreement, thus proving the feasibility of in-ear sensing for sleep monitoring in the community.

Introduction

Sleep quality is a fundamental pillar of human health, yet its accurate assessment remains a challenge. The current clinical gold standard, Polysomnography (PSG), is highly invasive, requiring patients to sleep in a laboratory setting while tethered to numerous electrodes. This "first-night effect" often distorts natural sleep patterns, limiting the ecological validity of the data. Conversely, consumer-grade wearables typically rely on actigraphy, which provides only an indirect proxy for sleep by measuring movement.

To bridge the gap between medical-grade precision and consumer-grade comfort, researchers have developed a proof-of-concept system for automatic sleep monitoring using a discreet, wearable in-ear EEG sensor (ear-EEG). This approach aims to provide direct neural monitoring in a form factor that is suitable for long-term, at-home use.

System Design and Methodology

The proposed system utilizes a viscoelastic in-ear sensor designed for comfort and signal stability. The sensor consists of a foam substrate that conforms to the ear canal, embedded with flexible cloth electrodes to capture the EEG signal.

The ear-EEG sensor design featuring a foam substrate and cloth electrodes, shown both as a model and in situ.

Feature Extraction and Classification

The core of the classification algorithm relies on two distinct types of features extracted from a single ear-EEG channel:

  1. Spectral Edge Frequency (SEF): A frequency-domain metric that identifies the frequency below which a specific percentage (e.g., 50% or 95%) of the total signal power resides. This captures the characteristic shift from high-frequency activity during wakefulness to low-frequency delta waves during deep sleep.
  2. Multi-Scale Fuzzy Entropy (MSFE): A measure of structural complexity. Unlike traditional spectral analysis, MSFE quantifies the predictability of the signal across multiple time scales. Brain activity typically becomes more deterministic and less complex as an individual transitions into deeper sleep stages (N3).

These features were processed using a Multi-class Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel. The study evaluated two primary scenarios:

  • Scenario 1: Predicting sleep labels derived directly from ear-EEG recordings.
  • Scenario 2: Predicting "gold standard" labels derived from traditional scalp-EEG recordings using only the ear-EEG data.

Experimental Protocol

The study involved four healthy male subjects (ages 25–36). Simultaneous recordings of four scalp-EEG channels and two ear-EEG channels were conducted during 45-minute sessions. To ensure a high density of state transitions, subjects were placed in a dark room where random 10-second noise bursts were played to induce shifts between wakefulness and sleep. Ground truth hypnograms were manually scored by a clinical expert following American Academy of Sleep Medicine (AASM) criteria.

Key Insights and Results

A critical finding of this research is the parity in signal quality between ear-EEG and scalp-EEG. Despite the smaller amplitude of signals recorded from the ear canal, the Signal-to-Noise Ratio (SNR) was found to be comparable to traditional scalp montages.

Power spectral density (PSD) comparison between scalp-EEG (C3-A2) and ear-EEG (EL1) across different sleep stages.

As shown in the spectral analysis, the ear-EEG (EL1) successfully captures the same fundamental physiological markers as the scalp-EEG (C3-A2). For instance, the alpha peak (~10 Hz) during wakefulness (W) is clearly visible in both, as is the shift toward lower frequencies in deeper sleep stages (N2, N3).

Classification Performance

The system demonstrated high accuracy across both 2-class (Wake vs. Sleep) and 4-class (W, N1, N2, N3) classification tasks:

  • 2-Class Accuracy: Reached 95.2% (kappa = 0.83) for Scenario 1 and 91.8% (kappa = 0.75) for Scenario 2.
  • 4-Class Accuracy: Reached 78.5% (kappa = 0.64) for Scenario 1 and 76.8% (kappa = 0.65) for Scenario 2.

The Cohen’s Kappa coefficients indicate "Substantial to Almost Perfect" agreement with expert scoring, validating the ear-EEG as a reliable surrogate for scalp-based monitoring.

Conclusion and Future Work

This preliminary study provides strong evidence that ear-EEG can serve as a viable, unobtrusive alternative for automatic sleep monitoring. The inclusion of non-linear complexity features like MSFE alongside traditional spectral metrics significantly enhances classification robustness.

While promising, the study's small sample size and short recording durations represent initial limitations. Future research will focus on validating the system over full 8-hour sleep cycles in a more diverse population, including individuals with sleep disorders. Additionally, the transition from laboratory amplifiers to fully integrated, wireless "hearable" hardware will be a key step toward widespread community adoption.

Summary updated

Abstract

The monitoring of sleep patterns without patient's inconvenience or involvement of a medical specialist is a clinical question of significant importance. To this end, we propose an automatic sleep stage monitoring system based on an affordable, unobtrusive, discreet, and long-term wearable in-ear sensor for recording the electroencephalogram (ear-EEG). The selected features for sleep pattern classification from a single ear-EEG channel include the spectral edge frequency and multi-scale fuzzy entropy, a structural complexity feature. In this preliminary study, the manually scored hypnograms from simultaneous scalp-EEG and ear-EEG recordings of four subjects are used as labels for two analysis scenarios: 1) classification of ear-EEG hypnogram labels from ear-EEG recordings; and 2) prediction of scalp-EEG hypnogram labels from ear-EEG recordings. We consider both 2-class and 4-class sleep scoring, with the achieved accuracies ranging from 78.5% to 95.2% for ear-EEG labels predicted from ear-EEG, and 76.8% to 91.8% for scalp-EEG labels predicted from ear-EEG. The corresponding Kappa coefficients range from 0.64 to 0.83 for Scenario 1, and indicate substantial to almost perfect agreement, while for Scenario 2 the range of 0.65-0.80 indicates substantial agreement, thus further supporting the feasibility of in-ear sensing for sleep monitoring in the community.
BibTeX
@article{nakamura2017automatic,
  author = {Takashi Nakamura and Valentin Goverdovsky and Mary J. Morrell and Danilo P. Mandic},
  title = {{Automatic Sleep Monitoring Using Ear-EEG}},
  journal = {IEEE Journal of Translational Engineering in Health and Medicine},
  year = {2017},
  doi = {10.1109/jtehm.2017.2702558},
}
Figure