InsightSleepNet: the interpretable and uncertainty-aware deep learning network for sleep staging using continuous Photoplethysmography

Table 2 Performance evaluation of the model on the MESA test dataset (n = 204) and a comparison before and after energy score thresholding

	InsightSleepNet (without thresholding)	0.80 Energy threshold	0.85 Energy threshold	0.90 Energy threshold	0.95 Energy threshold	BM-FE [19]	BM-DTS [19]	Sleep PPG-Net [19]
Accuracy	0.842	0.861	0.857	0.853	0.848	0.78	0.76	0.83
Cohen’s kappa (κ)	0.742	0.777	0.769	0.761	0.752	0.66	0.64	0.74
Weighted F1 score	0.8420	0.8613	0.8572	0.8528	0.8479	–	–	–

The ‘-’ symbol indicates that the corresponding metric is not provided in the study

ISSN: 1472-6947