SelANet: decision-assisting selective sleep apnea detection based on confidence score

Table 3 Performance comparison

Study	Dataset	Method	Signal	Acc (%)	AUROC	F1-score
Sharma et al.,2022 [43]	SHHS-1 (5,793)	Feature extraction + Decision tree	SpO2	79.81	N/A	0.792
			ECG	72.31	N/A	0.710
			SpO2 + ECG + AbdoRes + ThorRes	81.63	N/A	0.812
Pragya et al.,2022 [44]	SHHS-1 (5,793)	1D CNN	SpO2 + Pulse rate	84.3	0.862	N/A
Pragya et al.,2022 [44]	SHH-2 (2,651)	1D CNN	SpO2 + Pulse rate	82.2	0.904	N/A
Shanmugham et al.,2021 [21]	MGH	Feature extraction + ResNet	ECG + Respiration signal	77.00	0.840	N/A
Mahsa et al., 2021 [45]	Apnea-ECG (70)	LeNet + LSTM	ECG	80.67	N/A	0.747
Oliver et al., 2021 [46]	Apnea-ECG (70)	LSTM-CNN Hold out test	ECG (R-R interval)	81.30	85.32	N/A
Tom et al.,2018 [47]	SHHS-1 (5,793)	LSTM	ECG (EDR)	60.10	0.588	N/A
Tom et al.,2018 [47]	SHHS-1 (5,793)	LSTM	AbdoRes	77.20	0.775	N/A
This study (Without selective prediction)	MGH (994)	1D CNN-LSTM	ECG(EDR) + SaO2	83.22	0.908	0.832
This study (With selective prediction)	MGH (994)	1D CNN-LSTM + Selective prediction	ECG(EDR) + SaO2	90.26	0.939	0.905

Acc Accuracy, AUROC Area under the curve of the receiver operating characteristic, SHHS Sleep heart health study database, AbdoRes Abdomen respiration signal, ThorRes Thorax respiration signal, N/A Not applicable, MGH Massachusetts general hospital database, EDR ECG-Derived Respiration

ISSN: 1472-6947