Contacts: Dr. Amir Aminifar, Prof. David Atienza
External partners: SmartCardia SA
Motivation
Obstructive Sleep Apnea (OSA) is one of the most common sleep disorders, involving partial or complete obstruction of the upper airway. In the U.S. alone, an estimate of 3.8 million people between 30 and 60 years old are affected by this condition. Depending on the lifestyle, the prevalence of OSA could be up to 25%, with an estimate of 5% worldwide. This disorder is an aggravating factor for multiple health diseases, including cardiovascular ones (high blood pressure and stroke), clinical depression, as well as decreased memory and cognitive skills. In particular, previous studies demonstrate the high rate of sudden death of people with OSA because of cardiac causes. While this disorder is treatable, 90% of the subjects go undiagnosed.
Today, there is still a need to develop a non-intrusive solution for home OSA monitoring. On the one hand, there is low incentive for patients with low to moderate OSA to use external breathing equipment, such as Adaptive Servo Ventilation (ASV), because of its associated risks. In particular, a recent group study showed a higher mortality rate on OSA population with respiratory help compared to other patients. On the other hand, existing solutions are bulky, time-consuming, expensive, and intrusive. As the capacity of performing full OSA diagnosis does not match the recommended capacity, it makes the OSA testing and screening only available in dedicated facilities for the most severe cases: the patient is required to go to a sleep center or hospital where his or her sleep will be monitored extensively for two non-consecutive nights. A full polysomnography (PSG) will be done to record electrocardiogram (ECG), electromyogram (EMG), electroencephalogram (EEG), eye movements, nasal airflow. Altogether, it requires 22 electrodes plus a respiration mask. Moreover, the acquired data needs to be afterwards analyzed by a specialist. Given these constraints, more than 80% of patients are reluctant to undergo a PSG. In summary, the prevalence of OSA, the risks associated with external respiratory help, and the impracticality of current OSA screening motivate the need for a simple, yet efficient, solution for screening OSA.
Our Approach
In this project, we detect episodes of OSA events by monitoring a single-lead ECG signal on the INYU sensor, a wearable ECG monitoring device designed within the collaboration framework between the Embedded Systems Laboratory of EPFL and SmartCardia SA, which is shown below.
Considering the high rate of sudden death in people with OSA because of cardiac causes, in addition to OSA detection technique, it is also required to monitor the cardiovascular function (see here for details about real-time cardiac monitoring techniques). Therefore, the OSA detection technique has to be highly efficient to be able to monitor both OSA and cardiac function in real time and on a long-term basis. Our OSA detection technique is based on monitoring both cardiac and respiratory responses. It has a linear time-complexity on average, leveraging fast outlier removal, time-domain filtering and Parseval’s theorem, and linear Support Vector Machine (SVM) learning technique. Our lightweight OSA detection technique has a high classification performance of 88.2%, while having an autonomy of 46 days, measured experimentally on the INYU sensor. The overall flow of our proposed online OSA detection system is shown here and has three main entities: noise filtering, delineation, and OSA detection. In parallel, the raw signal is compressed and stored for further offline analysis.
