AI makes sleep apnea diagnosis simpler, cheaper, and home-based
Jiali Xie defended his PhD thesis at the Department of Electrical Engineering on May 22nd.
Better sleep could soon be within reach for millions, thanks to smart new tools that detect obstructive sleep apnea using nothing more than a heartbeat, a breath, or even a snore鈥攏o hospital stay required. Obstructive Sleep Apnea (OSA) is a common but often overlooked condition that affects between 6% and 17% of adults. It causes people to stop breathing repeatedly during sleep due to partial or complete blockage of the airway. The result? Poor sleep quality, daytime fatigue, and increased risk of serious health issues like heart disease, high blood pressure, and depression.
Why current tests fall short
Despite its impact, OSA is notoriously underdiagnosed鈥攍argely because the gold-standard test, known as polysomnography (PSG), requires an overnight stay in a sleep clinic connected to dozens of sensors. It鈥檚 accurate but expensive, uncomfortable, and not suited for large-scale screening or long-term monitoring.
Smarter tools from everyday signals
In his PhD research, developed machine learning models that can detect sleep apnea using simple, unobtrusive signals鈥攕uch as heart activity, breathing movement, and even snoring sounds. Many of these can be collected at home using wearable or bedside devices, opening the door to more accessible, affordable, and comfortable diagnostics.
From snores to insights
One part of the research focused on snoring鈥攐ne of the most common signs of sleep apnea. The researcher trained a deep learning model to recognize snoring sounds and used patterns in their rhythm and energy to estimate OSA severity. Another model analyzed heart signals (ECG) and breathing effort, distinguishing not just apneas but sleep from wakefulness鈥攃rucial for accurate diagnosis.
Combining data for better accuracy
The most advanced system combined audio and heart data into a single model. This multi-modal approach used the strengths of each type of signal, improving detection even when recordings were incomplete or noisy. In tests, combining signals consistently outperformed using any single type of data.
A future of at-home diagnosis
The bottom line? Sleep apnea can now be assessed reliably using tools that are low-cost, comfortable, and easy to use at home. These AI-powered systems offer new opportunities for early diagnosis, ongoing monitoring, and personalized treatment plans鈥攁ll without the burden of a sleep clinic visit.
Technology meets public health
This research isn鈥檛 just academic鈥攊t鈥檚 a real-world innovation that blends technology and healthcare in a way that could impact millions. By making sleep apnea diagnosis more accessible, it helps individuals take control of their health and could ease pressure on healthcare systems worldwide.
And all it takes is a heartbeat, a breath, or a snore.
Title of PhD thesis: . Supervisors: Prof. Sebastiaan Overeem, Dr. , and Dr. Hans van Dijk.