Table 1 Overview of validation studies classifying five stages

From: Evaluating the performance of wearable EEG sleep monitoring devices: a meta-analysis approach

Study	Device	# of al.^a	El. position	El. type	# of part.	Age	Part.	Env.^b	Nights	Ref.	Epochs	Device scoring^c		OA (mi)	OA (ma)	Wake	N1	N2	N3	REM
Forehead
Headband
Li et al. (2025)²⁷	WPSG-I²⁷	6	Fp1, Fp2, M1, M2, Chin 1 (EMG), Chin 2 (EMG)	Dry	20 (M: 13, F: 7)	56.2 ± 9.5	Healthy: 6, PD: 8, ALS: 5, nacrolepsy: 1	Controlled	1	PSG scored by 2 experts according to AASM	26,572	WPSG-I proprietary algorithm	ACC	0.89	0.96	0.96	0.93	0.93	0.98	0.98
													κ	0.80	0.71	0.91	0.32	0.79	0.82	0.70
													SE	0.89	0.74	0.97	0.34	0.83	0.93	0.63
													SP	0.97	0.97	0.93	0.97	0.96	0.98	0.99
													PPV	0.89	0.75	0.96	0.36	0.85	0.75	0.81
													NPV	0.97	0.97	0.95	0.96	0.95	1.00	0.98
													F1	0.89	0.74	0.97	0.35	0.84	0.83	0.71
													MCC	0.80	0.71	0.91	0.32	0.79	0.82	0.70
												Manual scoring	ACC	0.95	0.98	0.98	0.96	0.97	0.99	0.99
													κ	0.90	0.86	0.95	0.65	0.91	0.91	0.89
													SE	0.95	0.89	0.98	0.70	0.91	0.98	0.89
													SP	0.99	0.98	0.97	0.98	0.99	0.99	1.00
													PPV	0.95	0.86	0.98	0.63	0.95	0.86	0.89
													NPV	0.99	0.98	0.97	0.98	0.97	1.00	1.00
													F1	0.95	0.88	0.98	0.66	0.93	0.91	0.89
													MCC	0.90	0.86	0.95	0.65	0.91	0.91	0.89
Ravindran et al. (2025)²⁸	Dreem⁶⁹	5	F7, F8, O1, O2, Fpz	Dry	62 (M: 35, F: 27)	70.5 ± 6.7 (44–83)	Healthy: 50, AD: 12	Controlled	1	PSG scored by 2 experts according to AASM	63,546	Dreem proprietary algorithm	ACC	0.69	0.88	0.89	0.88	0.77	0.92	0.93
													κ	0.58	0.54	0.74	0.14	0.52	0.68	0.61
													SE	0.69	0.61	0.74	0.14	0.86	0.65	0.68
													SP	0.92	0.92	0.96	0.97	0.73	0.97	0.95
													PPV	0.69	0.65	0.89	0.34	0.59	0.83	0.62
													NPV	0.92	0.92	0.89	0.90	0.92	0.93	0.96
													F1	0.69	0.62	0.81	0.20	0.70	0.73	0.65
													MCC	0.59	0.55	0.74	0.16	0.54	0.69	0.61
Seol et al. (2024)²⁹	Insomnograf K2⁷³	4	Fp1, Fp2, M1, M2 (Fpz as ref)	Wet	77 (N/R)	>20 years	Suspected or known OSA	Controlled	1	PSG scored by an expert according to AASM	75,677	Manual scoring	ACC	0.78	0.91	0.96	0.84	0.85	0.96	0.96
													κ	0.71	0.73	0.86	0.55	0.69	0.71	0.81
													SE	0.78	0.78	0.92	0.56	0.90	0.67	0.85
													SP	0.95	0.94	0.97	0.94	0.82	0.99	0.97
													PPV	0.78	0.80	0.85	0.78	0.75	0.81	0.83
													NPV	0.95	0.94	0.99	0.85	0.93	0.97	0.98
													F1	0.78	0.78	0.88	0.65	0.82	0.73	0.84
													MCC	0.71	0.73	0.86	0.56	0.70	0.71	0.81
Rusanen et al. (2023)³⁰	Focusband⁷⁸	2	Fp1, Fp2 (Fpz as ref)	Dry	10 (M: 7, F: 3)	23–37	Healthy	Home (devices fitted by a specialist)	1	PSG scored by an expert according to AASM	9337	Deep learning (CNN)	ACC	0.82	0.93	0.94	0.98	0.85	0.93	0.94
													κ	0.75	0.67	0.78	0.26	0.69	0.79	0.67
													SE	0.82	0.70	0.80	0.21	0.87	0.77	0.86
													SP	0.96	0.95	0.97	0.99	0.83	0.98	0.96
													PPV	0.82	0.74	0.83	0.35	0.79	0.89	0.85
													NPV	0.96	0.95	0.96	0.98	0.90	0.94	0.97
													F1	0.82	0.72	0.82	0.28	0.82	0.83	0.85
													MCC	0.75	0.67	0.78	0.28	0.69	0.79	0.82
Casciola et al. (2021)³¹	Cognionics⁷⁹	4	F3, F4, A1, A2	Dry	12 (M: 6, F: 6)	21–61	Healthy	Controlled	1	PSG scored by an expert according to AASM	9747	Deep learning (CNN + LSTM)	ACC	0.74	0.90	0.92	0.89	0.83	0.95	0.90
													κ	0.64	0.59	0.76	0.20	0.65	0.76	0.57
													SE	0.74	0.69	0.82	0.28	0.74	0.87	0.73
													SP	0.93	0.93	0.94	0.93	0.91	0.96	0.92
													PPV	0.74	0.64	0.81	0.24	0.88	0.73	0.55
													NPV	0.93	0.93	0.95	0.95	0.79	0.98	0.96
													F1	0.74	0.66	0.81	0.26	0.80	0.79	0.63
													MCC	0.64	0.59	0.76	0.20	0.66	0.77	0.58
												Machine learning (Ensemble-bagged trees model)	ACC	0.68^d	0.49^d	N/R	N/R	N/R	N/R	N/R
												Machine learning (Ensemble-bagged trees model)	SE	N/R	N/R	0.80^d	0.04^d	0.82^d	0.50^d	0.28^d
Arnal et al. (2020)³²	Dreem⁶⁹	4	F7, F8, O1, O2, Fpz	Dry	25 (M: 19, F: 6)	35.3 ± 7.5 (23–50)	Mostly healthy, some with mild symptoms of anxiety or depression, one with insomnia	Controlled	1	PSG scored by 5 experts according to AASM	24,662	Deep learning (2 layers of LSTM + Softmax function)	ACC	0.81	0.92	0.95	0.93	0.86	0.95	0.93
													κ	0.73	0.69	0.76	0.43	0.72	0.78	0.78
													SE	0.81	0.76	0.78	0.48	0.83	0.87	0.86
													SP	0.95	0.95	0.97	0.96	0.90	0.96	0.95
													PPV	0.81	0.74	0.80	0.46	0.89	0.76	0.79
													NPV	0.95	0.94	0.97	0.96	0.84	0.98	0.97
													F1	0.81	0.75	0.79	0.47	0.86	0.81	0.83
													MCC	0.73	0.70	0.76	0.43	0.72	0.78	0.78
Lin et al. (2017)³³	Prototype developed in the study³³	4	AF7, Fp1, Fp2, AF8	Dry	10 (M: 10, F: 0)	24 ± 6	Healthy	Controlled	1	PSG scored by an expert according to AASM	8251	Machine learning (RVM)	ACC	0.77	0.91	0.94	0.86	0.86	0.95	0.92
													κ	0.69	0.65	0.80	0.23	0.70	0.83	0.72
													SE	0.77	0.72	0.84	0.22	0.85	0.87	0.83
													SP	0.94	0.94	0.96	0.96	0.86	0.97	0.94
													PPV	0.77	0.72	0.83	0.43	0.79	0.84	0.71
													NPV	0.94	0.94	0.96	0.89	0.90	0.98	0.97
													F1	0.77	0.71	0.83	0.29	0.82	0.86	0.77
													MCC	0.69	0.66	0.80	0.24	0.70	0.83	0.73
Levendowski et al. (2017)²²	X4 Sleep Profiler⁷⁰	2	AF7, AF8 (Fpz as ref)	Dry	47 (M: 35, F:12)	23–77	Sleep-disordered breathing and healthy	Controlled	1	PSG scored by 5 experts according to AASM	33,635	X4 Sleep Profiler proprietary algorithm	ACC	0.77	0.91	0.90	0.87	0.86	0.96	0.95
													κ	0.68	0.64	0.73	0.22	0.72	0.78	0.75
													SE	0.77	0.71	0.77	0.32	0.83	0.79	0.83
													SP	0.94	0.94	0.94	0.92	0.89	0.98	0.96
													PPV	0.77	0.70	0.82	0.26	0.85	0.82	0.73
													NPV	0.94	0.94	0.92	0.94	0.87	0.97	0.98
													F1	0.77	0.70	0.79	0.29	0.84	0.81	0.78
													MCC	0.68	0.64	0.73	0.22	0.72	0.78	0.75
												Automatic scoring corrected by reviewer	ACC	0.80	0.92	0.92	0.88	0.87	0.96	0.97
													κ	0.72	0.69	0.79	0.28	0.74	0.79	0.85
													SE	0.80	0.75	0.82	0.36	0.85	0.80	0.94
													SP	0.95	0.95	0.96	0.93	0.89	0.98	0.97
													PPV	0.80	0.74	0.87	0.33	0.86	0.83	0.80
													NPV	0.95	0.95	0.94	0.94	0.88	0.97	0.99
													F1	0.80	0.74	0.84	0.34	0.85	0.81	0.87
													MCC	0.72	0.69	0.79	0.28	0.74	0.79	0.85
Finan et al. (2016)³⁴	X4 Sleep Profiler⁷⁰	2	AF7, AF8 (Fpz as ref)	Dry	14 (M: 6, F: 8)	26.4 ± 3.7 (22–34)	Healthy	Controlled	1	PSG scored by an expert according to AASM	13,445*	X4 Sleep Profiler proprietary algorithm	ACC	0.66	0.86	0.92	0.89	0.78	0.88	0.85
													κ	0.52	0.45	0.42	0.07	0.55	0.62	0.58
													SE	0.66	0.55	0.44	0.27	0.72	0.60	0.72
													SP	0.92	0.91	0.96	0.91	0.83	0.96	0.89
													PPV	0.66	0.56	0.47	0.07	0.79	0.81	0.63
													NPV	0.92	0.91	0.96	0.98	0.77	0.89	0.92
													F1	0.66	0.54	0.46	0.11	0.75	0.69	0.67
													MCC	0.53	0.46	0.42	0.09	0.56	0.58	0.58
												Manual scoring	ACC	0.74	0.90	0.94	0.93	0.81	0.90	0.90
													κ	0.62	0.52	0.44	0.14	0.61	0.71	0.72
													SE	0.74	0.61	0.40	0.27	0.76	0.74	0.86
													SP	0.93	0.93	0.98	0.95	0.85	0.95	0.91
													PPV	0.74	0.61	0.57	0.12	0.81	0.80	0.72
													NPV	0.93	0.92	0.95	0.98	0.80	0.93	0.96
													F1	0.74	0.60	0.47	0.17	0.79	0.77	0.79
													MCC	0.62	0.53	0.45	0.15	0.61	0.71	0.73
Sleepmask
Liang et al. (2015)³⁵	Prototype developed in the study³⁵	2	EOG L, EOG R (Fpz as ref)	Dry	16 (M:11, F:5)	25.3 ± 2.5	Healthy	Controlled	1	PSG scored by an expert according to AASM	6480	Machine learning (LDA)	ACC	0.84	0.94	0.96	0.97	0.87	0.93	0.96
													κ	0.77	0.69	0.69	0.33	0.74	0.80	0.87
													SE	0.84	0.77	0.84	0.43	0.83	0.81	0.94
													SP	0.96	0.96	0.97	0.98	0.91	0.97	0.96
													PPV	0.84	0.71	0.62	0.29	0.89	0.87	0.87
													NPV	0.96	0.95	0.99	0.99	0.85	0.95	0.98
													F1	0.84	0.73	0.71	0.35	0.86	0.84	0.90
													MCC	0.77	0.69	0.70	0.34	0.74	0.80	0.88
Sheet-like/patches
Massie et al. (2025)³⁶	Prototype developed in the study³⁶	1	EOG R (Fpz as ref)	Dry	106 (M: 60, F: 46)	58 ± 15 (22–82)	Suspected OSA	Controlled	1	PSG scored by experts according to AASM	81,786	Deep learning (RNN)	ACC	0.80	0.92	0.92	0.95	0.84	0.94	0.95
													κ	0.70	0.66	0.77	0.44	0.69	0.62	0.78
													SE	0.80	0.74	0.84	0.47	0.82	0.76	0.80
													SP	0.95	0.94	0.94	0.97	0.88	0.95	0.98
													PPV	0.80	0.71	0.80	0.47	0.87	0.57	0.82
													NPV	0.95	0.94	0.95	0.97	0.82	0.98	0.97
													F1	0.80	0.72	0.82	0.47	0.84	0.65	0.81
													MCC	0.70	0.66	0.77	0.44	0.69	0.62	0.78
Roach et al. (2025)³⁷	Somfit⁸⁰	1	Fpz	Wet	27 (M: 13, F: 14)	22.3 ± 5.1	Healthy	Controlled	1	PSG scored by 3 experts according to AASM	21,600	Somfit proprietary algorithm	ACC	N/R	N/R	N/R	N/R	N/R	N/R	N/R
													κ	0.47^d	N/R	N/R	N/R	N/R	N/R	N/R
													SE	N/R	N/R	0.60^d	0.19^d	0.69^d	0.61^d	0.53^d
Um et al. (2025)³⁸	Prototype developed in the study³⁸	4	F7, F8, EOG L, EMG L, chin	Dry	1 (N/R)	N/R	Healthy	Controlled	1	PSG scored by an expert according to AASM	688	Deep learning (BiLSTM + attention model on spectrogram input)	ACC	0.73	0.89	0.96	0.78	0.79	0.98	0.95
													κ	0.61	0.65	0.74	0.42	0.59	0.66	0.83
													SE	0.73	0.75	0.72	0.73	0.70	0.77	0.83
													SP	0.93	0.93	0.98	0.79	0.90	0.99	0.98
													PPV	0.73	0.72	0.81	0.45	0.88	0.59	0.89
													NPV	0.93	0.92	0.98	0.93	0.74	0.99	0.96
													F1	0.73	0.72	0.76	0.56	0.78	0.67	0.86
													MCC	0.63	0.66	0.74	0.45	0.60	0.66	0.83
McMahon et al. (2024)³⁹	Somfit⁸⁰	1	Fpz	Wet	106 (M: 59, F: 47)	<65:85 ≥ 65:21	Suspected or known OSA	Controlled	1	PSG scored by 3 experts according to AASM	N/R	Deep learning (U-sleep CNN)	ACC	N/R	N/R	0.89^d	0.91^d	0.84^d	0.94^d	0.95^d
													κ	0.67^d	N/R	N/R	N/R	N/R	N/R	N/R
													SE	N/R	N/R	0.78^d	0.22^d	0.84^d	0.58^d	0.87^d
													SP	N/R	N/R	0.91^d	0.97^d	0.83^d	0.99^d	0.96^d
													PPV	N/R	N/R	0.76^d	0.38^d	0.76^d	0.85^d	0.76^d
													NPV	N/R	N/R	0.93^d	0.93^d	0.90^d	0.94^d	0.98^d
													F1	N/R	N/R	0.77^e	0.28^e	0.80^e	0.69^e	0.81^e
Oz et al. (2023)⁴⁰	X-trodes soft electrode array⁸¹	8	4 EEG (forehead), 2 EOG R, 2 EMG R (chin)	Dry	50 (M: 32, F: 18)	61.4 ± 7.9	Healthy: 21 PD: 29	Controlled	1	PSG scored by 2 experts according to the AASM	N/R	Manual scoring	ACC	0.77^d	N/R	N/R	N/R	N/R	N/R	N/R
													κ	0.69^d	N/R	0.70^d	0.22^d	0.58^d	0.41^d	0.72^d
													SE	N/R	N/R	0.91^d	0.16^d	0.84^d	0.68^d	0.77^d
													SP	N/R	N/R	0.94^d	0.99	0.80^d	0.97^d	0.98^d
													PPV	N/R	N/R	0.84^d	0.44^d	0.71^d	0.83^d	0.85^d
													F1	N/R	N/R	0.87^e	0.23^e	0.77^e	0.75^e	0.81^e
Kwon et al. (2023)⁴¹	Skin patch developed in the study⁴¹	5	2 EEG (forehead), EOG R, EOG L, EMG (chin)	Dry	8 (N/R)	N/R	Healthy	Controlled	1	PSG scored by an expert according to AASM	4961	Deep learning (CNN)	ACC	0.84	0.94	0.94	0.93	0.89	0.97	0.96
													κ	0.76	0.68	0.82	0.23	0.77	0.80	0.80
													SE	0.84	0.72	0.84	0.17	0.94	0.78	0.88
													SP	0.96	0.95	0.97	0.99	0.84	0.99	0.97
													PPV	0.84	0.78	0.88	0.52	0.85	0.86	0.77
													NPV	0.96	0.96	0.95	0.94	0.93	0.98	0.99
													F1	0.84	0.73	0.86	0.25	0.89	0.82	0.82
													MCC	0.76	0.69	0.82	0.27	0.78	0.81	0.80
												Manual scoring	ACC	0.82	0.93	0.92	0.90	0.88	0.96	0.98
													κ	0.74	0.69	0.76	0.23	0.76	0.78	0.92
													SE	0.82	0.73	0.72	0.30	0.94	0.75	0.92
													SP	0.96	0.95	0.98	0.94	0.83	0.99	0.99
													PPV	0.82	0.77	0.93	0.28	0.84	0.86	0.93
													NPV	0.96	0.95	0.92	0.95	0.93	0.97	0.99
													F1	0.82	0.74	0.81	0.29	0.89	0.80	0.93
													MCC	0.74	0.70	0.77	0.23	0.77	0.79	0.92
Matsumori et al. (2022)⁴²	Prototype developed in the study⁴²	6	Forehead (ref on mastoid)	Wet	27 (M:23, F:4)	27.4 ± 9.2	Healthy	Controlled	1	PSG scored by an expert according to the AASM	24,979	Deep learning (DSN: CNN + BiLSTM)	ACC	0.81	0.92	0.97	0.90	0.92	0.95	0.89
													κ	0.74	0.71	0.77	0.47	0.83	0.82	0.69
													SE	0.81	0.76	0.70	0.58	0.85	0.85	0.83
													SP	0.95	0.95	0.99	0.93	0.97	0.97	0.90
													PPV	0.81	0.77	0.88	0.48	0.95	0.86	0.70
													NPV	0.95	0.95	0.97	0.95	0.90	0.97	0.95
													F1	0.81	0.76	0.78	0.52	0.90	0.85	0.76
													MCC	0.74	0.72	0.77	0.47	0.83	0.82	0.69
Myllymaa et al. (2016)⁴³	Bittium Brainstatus EEG⁸²	11	Fp1, Fp2, AF7, AF8, F8, F7, Sp1, Sp2, T10, T9, EOG R	Wet	31 (M:10, F: 21)	31.3 ± 11.8	Sleep bruxism or healthy	Controlled	1	PSG scored by 2 experts according to the AASM	27,692	Manual scoring	ACC	0.80	0.92	0.94	0.89	0.86	0.94	0.96
													κ	0.71	0.69	0.76	0.43	0.72	0.79	0.77
													SE	0.80	0.74	0.76	0.57	0.87	0.77	0.75
													SP	0.95	0.94	0.97	0.93	0.85	0.98	0.98
													PPV	0.80	0.77	0.83	0.44	0.83	0.89	0.85
													NPV	0.95	0.94	0.96	0.96	0.89	0.95	0.97
													F1	0.80	0.75	0.79	0.49	0.85	0.83	0.79
													MCC	0.71	0.70	0.76	0.44	0.72	0.79	0.77
Ear
In-ear
Borges et al. (2025)⁴⁴	Prototype developed in Goverdovsky et al. (2016⁷⁴, 2017)⁷⁵	4	2 per ear	Wet	14 (M: 6, F: 8)	53.2 ± 17.4 (25–78)	Mostly OSA	Controlled	1	PSG scored by an expert according to AASM	30,960	Automatic (not specified)	ACC	0.83	0.93	0.94	0.91	0.89	0.95	0.98
													κ	0.77	0.77	0.83	0.56	0.77	0.77	0.89
													SE	0.83	0.80	0.81	0.60	0.91	0.74	0.93
													SP	0.96	0.95	0.98	0.95	0.87	0.98	0.98
													PPV	0.83	0.83	0.93	0.62	0.82	0.87	0.88
													NPV	0.96	0.96	0.95	0.94	0.94	0.96	0.99
													F1	0.83	0.81	0.87	0.61	0.86	0.80	0.91
													MCC	0.77	0.77	0.83	0.56	0.77	0.78	0.89
Hammour et al. (2024)⁴⁵	Prototype developed in Goverdovsky et al. (2016)⁷⁴, (2017)⁷⁵	4	2 per ear	Wet	13 (M: 9, F: 4)	71.8 ± 4.4 (65–83)	Mostly healthy; some with stable comorbidities (type-2 diabetes, sleep apnea, hypertension)	Controlled	1	PSG scored by 2 scorers according to AASM	13,403	Machine learning (fine-tuned pre-trained LightGBM⁸³)	ACC	0.74	0.90	0.90	0.86	0.83	0.94	0.95
													κ	0.64	0.62	0.79	0.29	0.59	0.73	0.69
													SE	0.74	0.67	0.90	0.36	0.72	0.74	0.64
													SP	0.93	0.93	0.90	0.92	0.87	0.97	0.98
													PPV	0.74	0.71	0.84	0.38	0.70	0.79	0.81
													NPV	0.93	0.93	0.94	0.91	0.88	0.96	0.96
													F1	0.74	0.69	0.87	0.37	0.71	0.77	0.72
													MCC	0.64	0.62	0.79	0.29	0.59	0.73	0.70
Borup et al. (2023)⁴⁶	Prototype developed in Kappel et al. (2018)⁷²	12	6 per ear	Dry	20 (M: 7, F:13)	25.9 (22–36)	Healthy	Home (devices fitted by a specialist)	4	PSG scored by an expert according to AASM	72,942	Deep learning (personalized ensemble deep learning)	ACC	0.84	0.94	0.98	0.94	0.88	0.95	0.94
													κ	0.78	0.75	0.88	0.47	0.76	0.85	0.80
													SE	0.84	0.79	0.92	0.44	0.89	0.83	0.85
													SP	0.96	0.96	0.98	0.98	0.87	0.99	0.96
													PPV	0.84	0.81	0.88	0.57	0.83	0.95	0.82
													NPV	0.96	0.96	0.99	0.96	0.92	0.95	0.97
													F1	0.84	0.80	0.90	0.50	0.86	0.88	0.84
													MCC	0.78	0.75	0.88	0.47	0.76	0.86	0.80
Tabar et al. (2023)⁴⁷	Prototype developed in the study⁴⁷	4	2 per ear	Dry	10 (M: 6, F: 4)	27.4 ± 4.9 (22–35)	Healthy	Home (device fitted by participants, PSG by specialist)	2	Partial PSG scored by an expert according to AASM	15,709	Machine learning (Random Forest)	ACC	0.81	0.92	0.96	0.92	0.85	0.96	0.93
													κ	0.72	0.65	0.70	0.23	0.70	0.86	0.77
													SE	0.81	0.71	0.76	0.17	0.89	0.90	0.81
													SP	0.95	0.94	0.97	0.99	0.81	0.97	0.96
													PPV	0.81	0.75	0.70	0.55	0.81	0.87	0.82
													NPV	0.95	0.95	0.98	0.93	0.89	0.98	0.95
													F1	0.81	0.71	0.73	0.25	0.85	0.89	0.81
													MCC	0.72	0.66	0.70	0.27	0.71	0.86	0.77
Jørgensen et al. (2023)⁴⁸	Prototype developed in Kappel et al. (2018)⁷²	12	6 per ear	Dry	1 (M: 0, F:1)	29 ± 3.8 (22–35)	Healthy	Home (devices fitted by a specialist)	2	PSG scored by an expert according to AASM	1578	Machine learning (Random Forest)	ACC	0.79	0.92	0.98	0.91	0.82	0.93	0.95
													κ	0.72	0.67	0.83	0.24	0.60	0.81	0.88
													SE	0.79	0.71	0.90	0.48	0.58	1.00	0.96
													SP	0.95	0.95	0.99	0.93	0.98	0.91	0.95
													PPV	0.79	0.78	0.78	0.20	0.94	0.75	0.89
													NPV	0.95	0.95	1.00	0.98	0.77	1.00	0.98
													F1	0.79	0.72	0.84	0.28	0.72	0.86	0.92
													MCC	0.73	0.69	0.83	0.27	0.63	0.83	0.89
Kjaer et al. (2022)⁴⁹	Prototype developed in Kappel et al. (2018)⁷²	12	6 per ear	Dry	20 (M: 7, F: 13)	25.9 ± 3.8 (22–36)	Healthy	Home (devices fitted by a specialist)	4	PSG scored by 2 experts according to AASM	72,942	Machine learning (Random Forest)	ACC	0.80	0.92	0.96	0.94	0.84	0.96	0.91
													κ	0.73	0.65	0.84	0.15	0.68	0.85	0.70
													SE	0.80	0.70	0.91	0.10	0.87	0.85	0.75
													SP	0.95	0.94	0.97	1.00	0.82	0.98	0.95
													PPV	0.80	0.76	0.82	0.54	0.78	0.92	0.76
													NPV	0.95	0.95	0.98	0.95	0.90	0.96	0.94
													F1	0.80	0.70	0.87	0.16	0.82	0.88	0.76
													MCC	0.73	0.66	0.84	0.21	0.68	0.86	0.70
Jørgensen et al. (2020)⁵⁰	Prototype developed in Zibrandtsen et al. (2016)⁸⁴	8	4 per ear	Wet	13 (M: 5, F: 8)	41.5 (18–60)	Epilepsy	Controlled	1–4	PSG scored by an expert according to AASM	27,593	Manual scoring	ACC	0.81	0.92	0.96	0.89	0.85	0.94	0.98
													κ	0.74	0.74	0.85	0.47	0.70	0.79	0.90
													SE	0.81	0.80	0.91	0.63	0.80	0.80	0.87
													SP	0.95	0.95	0.97	0.92	0.89	0.97	0.99
													PPV	0.81	0.79	0.84	0.46	0.86	0.86	0.96
													NPV	0.95	0.95	0.98	0.96	0.85	0.96	0.98
													F1	0.81	0.79	0.87	0.53	0.83	0.83	0.91
													MCC	0.74	0.74	0.85	0.48	0.70	0.80	0.90
Nakamura et al. (2020)⁵¹	Prototype developed in Goverdovsky et al. (2016)⁷⁴, (2017)⁷⁵	4	2 per ear	Wet	22 (N/R)	23.8 ± 4.8	Healthy	Home (devices fitted by specialist)	1	PSG scored by an expert according to AASM	11,610	Machine learning (SVM)	ACC	0.74	0.90	0.89	0.99	0.77	0.93	0.90
													κ	0.61	0.51	0.69	0.09	0.55	0.75	0.49
													SE	0.74	0.57	0.74	0.05	0.84	0.75	0.46
													SP	0.94	0.92	0.94	1.00	0.71	0.97	0.97
													PPV	0.74	0.73	0.77	0.64	0.71	0.84	0.67
													NPV	0.94	0.93	0.93	0.99	0.84	0.95	0.92
													F1	0.74	0.59	0.76	0.09	0.77	0.79	0.54
													MCC	0.62	0.53	0.69	0.18	0.56	0.75	0.50
Mikkelsen et al. (2019)⁵²	Prototype developed in Kappel et al. (2018)⁷²	12	6 per ear	Dry	20 (M:7, F:13)	25.9 (22–36)	Healthy	Home (devices fitted by a specialist)	4	Partial PSG scored by an expert according to AASM	72,942	Machine learning (Random Forest)	ACC	0.81	0.92	0.96	0.93	0.85	0.96	0.91
													κ	0.73	0.68	0.85	0.29	0.69	0.86	0.70
													SE	0.81	0.77	0.84	0.52	0.79	0.93	0.77
													SP	0.95	0.95	0.98	0.94	0.90	0.97	0.94
													PPV	0.81	0.72	0.91	0.23	0.88	0.85	0.74
													NPV	0.95	0.94	0.97	0.98	0.82	0.99	0.95
													F1	0.81	0.73	0.87	0.32	0.83	0.89	0.76
													MCC	0.73	0.68	0.85	0.31	0.69	0.86	0.70
Mikkelsen et al. (2017)⁵³	Prototype developed in the study⁵³	12	6 per ear	Wet	9 (M: 6, F: 3)	26–44	Healthy	Home (devices fitted by a specialist)	1	Partial PSG scored by experts according to AASM	7411	Machine learning (Random Forest)	ACC	0.60	0.84	0.84	0.92	0.74	0.91	0.80
													κ	0.45	0.40	0.52	0.04	0.47	0.61	0.34
													SE	0.60	0.52	0.53	0.16	0.69	0.74	0.46
													SP	0.90	0.89	0.94	0.93	0.78	0.94	0.88
													PPV	0.60	0.51	0.74	0.04	0.71	0.59	0.47
													NPV	0.90	0.89	0.86	0.98	0.77	0.97	0.87
													F1	0.60	0.50	0.61	0.07	0.70	0.66	0.47
													MCC	0.45	0.40	0.53	0.05	0.47	0.61	0.34
Around-the-ear
da Silva Suoto et al. (2022)⁵⁵	Prototype developed in the study⁵⁵	7	2 EMG, 1 EOG, 2 forehead, 2 around-the-ear	Wet	12 (M: 9, F: 3)	28.9 (18–45)	Healthy	Home (devices fitted by a specialist)	1	PSG scored by an expert according to AASM	10,632	Manual scoring	ACC	0.78	0.93	0.95	0.91	0.85	0.91	0.93
													κ	0.70	0.57	0.75	0.46	0.66	0.79	0.76
													SE	0.78	0.62	0.84	0.56	0.79	0.83	0.72
													SP	0.96	0.95	0.96	0.94	0.87	0.95	0.98
													PPV	0.78	0.62	0.72	0.46	0.76	0.87	0.90
													NPV	0.96	0.95	0.98	0.96	0.89	0.93	0.94
													F1	0.78	N/R	0.77	0.51	0.77	0.85	0.80
													MCC	0.70	0.57	0.75	0.46	0.66	0.79	0.77
da Silva Suoto et al. (2021)⁵⁴	cEEGrid⁸⁵	16	8 per ear	Wet	10 (M: 2, F: 8)	28.4 ± 4.3	Healthy	Home (devices fitted by a specialist)	1	EEG of Fpz, EOG_L and EOG_R scored by expert according to AASM	9341	Manual scoring	ACC	0.75	0.92	0.94	0.92	0.82	0.95	0.91
													κ	0.67	0.60	0.71	0.37	0.62	0.85	0.69
													SE	0.75	0.70	0.69	0.35	0.82	0.90	0.67
													SP	0.95	0.94	0.98	0.97	0.82	0.97	0.97
													PPV	0.75	0.67	0.80	0.50	0.74	0.87	0.83
													NPV	0.95	0.95	0.96	0.95	0.88	0.97	0.93
													F1	0.75	0.65	0.74	0.41	0.78	0.88	0.74
													MCC	0.67	0.61	0.71	0.38	0.62	0.85	0.70
Mikkelsen et al. (2019)⁵⁶	cEEGrid⁸⁵	16	8 per ear	Wet	15 (M: 6, F: 9)	35.3 ± 14.3	Healthy	Controlled	1	PSG scored by 2 experts according to AASM	18,920	Machine learning (Random Forest)	ACC	0.700^f	N/R	N/R	N/R	N/R	N/R	N/R
Mikkelsen et al. (2019)⁵⁶	cEEGrid⁸⁵	16	8 per ear	Wet	15 (M: 6, F: 9)	35.3 ± 14.3	Healthy	Controlled	1	PSG scored by 2 experts according to AASM	18,920	Machine learning (Random Forest)	κ	0.600^f	N/R	N/R	N/R	N/R	N/R	N/R
Sterr et al. (2018)⁵⁷	cEEGrid⁸⁵	16	8 per ear	Wet	15 (M: 6, F: 9)	35.3 ± 14.3	Healthy	Controlled	1	PSG scored by 2 experts according to AASM	18,920	Manual scoring	ACC	0.59^d	N/R	N/R	N/R	N/R	N/R	N/R
Sterr et al. (2018)⁵⁷	cEEGrid⁸⁵	16	8 per ear	Wet	15 (M: 6, F: 9)	35.3 ± 14.3	Healthy	Controlled	1	PSG scored by 2 experts according to AASM	18,920	Manual scoring	κ	0.42^d	N/R	N/R	N/R	N/R	N/R	N/R

All evaluation metric values were calculated from confusion matrices provided in the original publications unless indicated otherwise.
AASM American Association of Sleep Medicine, ACC accuracy, AD Alzheimer’s disease, ALS amyotrophic lateral sclerosis, BiLSTM bidirectional long short-term memory, CNN convolutional neural network, DSN deep stacking networks, EEG electroencephalogram, F females, F1 F1 score; κ Cohen’s kappa, LDA linear discriminant analysis; LightGBM: light gradient boosting machine; LSTM: long short-term memory, M males, MCC Matthews correlation coefficient, NPV negative predictive value, OA (ma) overall macro-averaged metrics, OA (mi) overall micro-averaged metrics; OSA: obstructive sleep apnea, PD Parkinson’s disease, PPV positive predictive value, REM rapid eye movement, RNN recurrent neural network, RVM relevance vector machine, SE sensitivity, SP specificity, SVM support vector machine, WPSG-I wearable polysomnogram, N/R not reported.
^aThe number of recording electrodes.
^bFor controlled environment studies, devices were fitted by researchers or technicians. For home studies, self-application vs. expert fitting is indicated in parentheses.
^cIn studies reporting manual scoring of wEEG data, scoring was conducted on raw signals from the device. As most wEEG devices lacked EOG and EMG channels, REM sleep was identified based on EEG characteristics alone, such as low-amplitude mixed-frequency activity.
^dReported in the original study.
^eCalculated from reported metrics.
^fEstimated from the graph.

Back to article page

Table 1 Overview of validation studies classifying five stages

Search

Quick links