Towards detection of sleep apnea events by combining different non-contact measurement modalities.

In this work, we extract features from an under-the-mattress impulse radio ultra-wide band (IR-UWB) radar and a microphone, placed on the side table of the bed, to classify epochs belonging to normal sleep and those that contain an apnea event in them. Sleep apnea is the most common form of sleep related breathing disorder in adults with an estimated prevalance of 5-15%. The common diagnostic process for sleep apnea, polysomnography (PSG), involves sleeping in well-equipped sleep clinics. The cost and discomfort associated with the process has spurred research towards the design of portable home-based monitoring devices. However, these include sensors which need to be attached to patients at various locations on the body. In this preliminary and on-going study, we collected 18 hours of data of 3 subjects who were previously diagnosed with sleep apnea. The data was recorded using non-contact sensors, an IR-UWB radar and a microphone, in a sleep clinic along with the time synchronized gold-standard PSG data. A simple linear classifier was used to perform binary classification between normal and apnea epochs and the performance was analyzed compared to the true results provided by the PSG. It was observed, that combining snore features from the microphone data improves the overall accuracy of the classifier.