Contactless Oxygen Monitoring with Radio Waves and Gated Transformer

With the increasing popularity of telehealth, it is crucial to ensure accurate monitoring of basic physiological signals at home with minimal patient overhead. In this paper, we propose a contactless approach for monitoring blood oxygen levels simply by analyzing radio signals in a patient’s room, without the need for wearable devices. Our method extracts a patient’s respiration from radio signals that bounce off their body, and we use a novel neural network, called Gated BERT-UNet, to estimate blood oxygen saturation from the breathing signal. We designed our model to adapt to a patient’s medical indices, such as gender and sleep stages, to provide personalized inference. Specifically, it uses multiple predictive heads, controlled by a gate, to make predictions for different sub-populations. Our extensive empirical results demonstrate that our model achieves high accuracy on both medical and radio-frequency datasets. It outperforms past work on contactless oxygen monitoring, reducing the mean absolute error in oxygen saturation from 2.0% to 1.3%.