Author(s): Borges, Daniel Filipe ; Soares, Joana I. ; Silva, Heloísa ; Felgueiras, João ; Batista, Carla ; Ferreira, Simão ; Rocha, Nuno ; Leal, Alberto
Date: 2024
Persistent ID: http://hdl.handle.net/10400.22/26270
Origin: Repositório Científico do Instituto Politécnico do Porto
Subject(s): Sleep; Polysomnography
Introduction:Sleep is vital for health. It has regenerative and protective functions, and its disruption reduces the quality of life and increases susceptibility to disease. During sleep, there is a cyclicity of distinct phases that are studied using polysomnography (PSG), a costly and technically demanding method that compromises the quality of natural sleep. The search for simpler devices for recording biological signals at home addresses some of these issues. Objective: To clinically validate a custom-built single-channel in-ear EEG sensor for sleep classification by assessing various sleep metrics and staging decisions with simultaneously recorded PSG. Methods: Prospective cross-sectional study with 28 participants, divided into two groups: healthy volunteers and clinical patients. In both groups, PSG, individual in-ear EEG- with two different electrode configurations- and actigraphic recordings (only in the healthy group) were performed simultaneously for a whole night. Statistical analysis focussed on the four main sleep metrics: TRT (total recording time), TST (total sleep time), SE (sleep efficiency), SL (sleep latency) and the 5-class classifications (wakefulness, N1, N2, N3 and REM sleep). This included correlation analyses between methods and Bland-Altman plots, Cohen’s K coefficient, and confusion matrices aiming 30-second epoch-wise agreement with an automatic sleep classification algorithm using visual sleep classification by an ERSR-certified human expert as the gold standard according to current AASM guidelines. Results: The analysed sleep data comprised 30960 epochs. The correlation analysis revealed strong positive correlations (0.90) for all variables for the in-ear sensor. The Bland-Altman plots show a high level of agreement and consistency (+- 1.87 SD), with minimal bias between methods. The average kappa values (0.75) and the confusion matrices with each method's sensitivity and specificity also show a very high level of concordance.Conclusions: In both groups, the in-ear EEG sensor showed strong correlation, agreement and reliability with the gold standard, supporting accurate sleep classification.
