A Novel Oxygen-Hemoglobin Model for Non-Contact Sleep Monitoring of Oxygen Saturation

This paper proposes a novel mathematical model that quantitatively defines the relationships between the arterial oxygen saturation (SaO2), the arterial partial pressure of oxygen (PaO2) and the arterial partial pressure of carbon dioxide (PaCO2). A novel non-contact algorithm that utilizes the mathematical model, multilayer perceptron (MLP) neural network and microwave Doppler radar to translate the human periodic chest displacements caused by respiratory efforts into peripheral capillary oxygen saturation (SpO2) measurements is also proposed. A database consisting of 20 obstructive sleep apnea (OSA) and chronic heart failure (CHF) patients, with total sleep duration of 155 hours, 1 minute and 30 seconds, is selected for the “ Training,” “ Validation ” and independent “ Test ” of the SpO2 predictions. For independent “ Test ” dataset with total sleep duration of 54 hours, 15 minutes and 31.5 seconds, the SpO2 predictions correlation coefficient achieved 0.93 with sum squared error (SSE) of 1.3 (% of oxygen saturation) and the 95% limits of agreement of ±2.5 (% of oxygen saturation). These results demonstrate and proven the feasibility of non-contact SpO2 predictions using microwave Doppler radar in the complexity of sleep environment. A potential application could be non-contact sleep monitoring of oxygen saturation in the home.