Sleep apnoea, a condition marked by pauses in breathing during sleep, affects millions of people worldwide, leading to severe health risks, including heart disease, stroke, and diabetes. Traditionally, diagnosing this condition requires an overnight stay at a clinic for polysomnography (PSG) tests, which are costly, complex, and uncomfortable as they involve attaching multiple sensors to the body, making it difficult to sleep naturally. In addition, limited availability of sleep clinics can lead to long waiting times for appointments which can delay the diagnosis. But what if there was a way to monitor and diagnose sleep apnoea from the comfort of your own bed? The team of Prof. Jorge Gonçalves aims to revolutionise how sleep apnoea is diagnosed by enabling patients to monitor their sleep at home using simple, comfortable sensors.
By analysing more than 14,000 sleep study recordings from the National Sleep Research Resource , the researchers from the Luxembourg Centre for Systems Biomedicine (LCSB) at the University of Luxembourg developed DRIVEN, a breakthrough method that uses wearable sensors and artificial intelligence (AI) to estimate the severity of apnoea without the need for elaborate equipment and overnight stay at the clinic. The study was recently published in npj Digital Medicine , one of the leading scientific journals in this field, and highlights DRIVEN’s potential to make apnoea diagnosis more accessible and reduce the burden on healthcare systems.
AI in action for sleep analysis
The DRIVEN method uses artificial intelligence to analyse just a few signals - like abdominal movement, thoracic movement, and pulse oximetry - captured by easy to wear sensors. These sensors can be used at home and provide enough data to estimate the Apnoea-Hypopnea Index (AHI), a key metric used by doctors to determine the severity of sleep apnoea consisting of the number of apnoea events per hour. The AHI index is divided into four categories: healthy, mild, moderate and severe. The researchers compared the AHI score generated by DRIVEN to the gold standard in which a doctor analyses the polysomnography data from the sleep lab. "Based on data from only a single night of sleep, using just two sensors, we were able to place 99.3% of patients either in the correct AHI category (72.4%) or one class away from the true one (26,9%)," explains Gabriela Retamales , PhD candidate in the Systems Control group at the University’s LCSB. "If people were to wear those sensors for several nights in a row, we expect the accuracy to further improve."
Why are wearable sensors a game-changer?
Traditional PSG tests involve attaching around 25 sensors with cables to the body, making it difficult for patients to sleep naturally.
DRIVEN, on the other hand, achieves high accuracy with minimal equipment.
The researchers found that the best combination involved using an abdominal movement sensor and pulse oximetry, allowing DRIVEN to measure breathing patterns and oxygen levels with minimal intrusion.
"There are no cables. Pulse oximetry can be measured with a smartwatch or with a sensor on a finger, while abdominal movement can be captured with an elastic around the waist and a small device to store the data. The data is then download in the morning. By focusing on a few key signals, we can hence offer patients a lot more comfort while still providing accurate information for doctors,- explains Prof. Jorge Gonçalves, senior author of the study and principal investigator of the Systems Control group.
Fast and accurate results with AI
The real innovation behind DRIVEN is the use of artificial intelligence to analyse sleep data. It provides an end-to-end solution, ready to be deployed. Conventional methods rely on manual feature extraction by experts: a time-consuming and often subjective process which is also influenced by the total amount of sleep during the test. DRIVEN automates all these steps through deep convolutional neural networks, which "learn- from large datasets and extract meaningful patterns in breathing, oxygen levels and consider the total amount of sleep. This not only speeds up the diagnostic process but also eliminates the human error that can be associated with traditional methods.
A new chapter for sleep medicine
DRIVEN’s potential extends beyond just offering convenience. By enabling long-term, unsupervised monitoring of sleep patterns at home, it could significantly reduce the costs of diagnosing sleep apnoea and improve patient outcomes. "This system could become an affordable and effective solution for millions of undiagnosed patients,- concludes Prof. Jorge Gonçalves. "While my research group primarily focuses on fundamental questions, I am eager to collaborate with clinical and industrial partners to turn this idea into a marketable medical product. Moving from theory to innovation could have a profound impact on patient care and quality of life.- As wearable health technology continues to evolve, the future of sleep medicine may lie in the hands - or on the wrists - of patients themselves.
Original publication:
Retamales, G., Gavidia, M.E., Bausch, B. et al. Towards automatic home-based sleep apnea estimation using deep learning. npj Digit. Med. 7, 144 (2024).
Prof Jorge GONCALVES
Full professor / Chief scientist 1 in Computational Biology
