Comparison analysis between standard polysomnographic data and in-ear-electroencephalography signals: a preliminary study

Abstract

Study objectives: Polysomnography (PSG) currently serves as the benchmark for evaluating sleep disorders. Its discomfort makes long-term monitoring unfeasible, leading to bias in sleep quality assessment. Hence, less invasive, cost-effective, and portable alternatives need to be explored. One promising contender is the in-ear-electroencephalography (EEG) sensor. This study aims to establish a methodology to assess the similarity between the single-channel in-ear-EEG and standard PSG derivations.

Methods: The study involves 4-hour signals recorded from 10 healthy subjects aged 18-60 years. Recordings are analyzed following two complementary approaches: (1) a hypnogram-based analysis aimed at assessing the agreement between PSG and in-ear-EEG-derived hypnograms; and (2) a feature- and analysis-based on time- and frequency-domain feature extraction, unsupervised feature selection, and definition of Feature-based Similarity Index via Jensen-Shannon Divergence (JSD-FSI).

Results: We find large variability between PSG and in-ear-EEG hypnograms scored by the same sleep expert according to Cohen's kappa metric, with significantly greater agreements for PSG scorers than for in-ear-EEG scorers (p < .001) based on Fleiss' kappa metric. On average, we demonstrate a high similarity between PSG and in-ear-EEG signals in terms of JSD-FSI-0.79 ± 0.06-awake, 0.77 ± 0.07-nonrapid eye movement, and 0.67 ± 0.10-rapid eye movement-and in line with the similarity values computed independently on standard PSG channel combinations.

Conclusions: In-ear-EEG is a valuable solution for home-based sleep monitoring; however, further studies with a larger and more heterogeneous dataset are needed.

Keywords: in-ear-EEG; machine learning; multisource-scored sleep databases; sleep staging; sleep wearables.

Figure 1.

Schematic layout of the quality assurance study and the data collection procedure.

Figure 2.

Devices employed in the data collection: (A) SOMNOmedics SOMNOscreen plus system for PSG data with the EXG configuration, that is, including six scalp electrodes, EOG, and ECG signal monitoring; (B) GDK hardware including ear tips, earpieces, and brain box used to record in-ear-EEG data.

Figure 3.

Workflow for evaluating the similarity between the signals recorded from two different channels, including feature extraction and feature selection, separately for each sleep stage; and the comparison between feature distributions using the Jensen–Shannon divergence before the assessment of the similarity-scores, individually for each sleep stage and for each subject. In detail, (A) refers to the comparison between one in-ear-EEG and one PSG channel (either scalp-EEG or EOG channels); while (B) illustrates the analysis between two PSG channels (either scalp-EEG or EOG channels). An example of similarity-scores distribution for awake, NREM, and REM classes is included for both case studies.

Figure 4.

(A) Intrascorer variability (multisource-scored dataset). Boxplot distribution of the Cohen’s kappa values computed for each recording/subject between the PSG and in-ear-EEG hypnograms—for each scorer. (B) Interscorer variability (multisource-scored dataset). Boxplot distributions of the Fleiss’ kappa values computed for each recording/subject between the three scorer experts for in-ear-EEG (in blue) and PSG (in red) signals.

Figure 5.

Example of 30-second data samples from both PSG and in-ear EEG recordings, for each sleep stage. The PSG derivation considered is the C3-M2 channel.

Figure 6.

JSD-FSI similarity scores distributions, that is, distributions derived from the PSG-to-in-ear-EEG (histograms in blue) and PSG-to-PSG (histograms in red) comparisons—for each subject in the awake stage.

Figure 7.

JSD-FSI similarity-scores distributions, that is, distributions derived from the PSG-to-in-ear-EEG (histograms in blue) and PSG-to-PSG (histograms in red) comparisons—for each subject in the NREM sleep stage.

Figure 8.

JSD-FSI similarity-scores distributions, that is distributions derived from the PSG-to-in-ear-EEG (histograms in blue) and PSG-to-PSG (histograms in red) comparisons—for each subject in the REM sleep stage. No JSD-FSI similarity scores are reported for subjects 3 and 8.