Umaer Hanif_Søvnapnø

Artificial intelligence can diagnose sleep apnoea

Thursday 02 Dec 21

Researchers from DTU Health Tech have developed a new 3D scanning based screening method for diagnosing sleep apnoea.

Obstructive sleep apnoea is a disease, where the upper part of the airways is blocked for more than 10 seconds during sleep. This means that sleep and ultimately the quality of a person’s sleep is disturbed, when he is choked repeatedly throughout the night. Some studies suggest that as many as 20% of the population suffer from sleep apnoea. However, the numbers vary as the disorder is underdiagnosed. A lot of people are simply not aware that they have the disease, because the symptoms, which includes tiredness and feeling off one’s game during the day, are diffuse and can be attributed to a number of things besides sleep apnoea.

PhD student Umaer Rashid Hanif has developed a new screening technology for diagnosing obstructive sleep apnoea. Umaer is part of Associate Professor Helge B.D. Sørensen’s research group, Biomedical Signal Processing & AI, at DTU Health Tech. On the basis of a 3D scanning of a patient’s face and neck region, he can predict if the patient suffers from obstructive sleep apnoea and if he or she should be sent for an actual sleep test in a sleep lab.

“The project is a collaboration between DTU Health Tech, Rigshospitalet Glostrup (via Prof. Jennum’s group) and Stanford University (Prof. Mignot’s group), Center for Sleep Sciences and Medicine, where I have spent a year and a half of my PhD. The researchers at Stanford also did the 3D scans, which I have worked with. It is common knowledge among sleep experts that a person’s facial anatomy and neck area, for example if the patient has a lot of fat around the neck, are correlated with sleep apnoea. My part of the project was to design and train a suitable algorithm. That is, to use artificial intelligence and machine learning to train a model based on the collected data”, Umaer Rashid Hanif explains.

Data driven method equal to current screening tool

The method used for developing and training the artificial intelligence is entirely data driven. This means that the researchers only feed the model with data. In this case data consisted of 3D scans of patients, who came in for a sleep test, as well as the results from their actual sleep tests, where the patients’ breathing were monitored by sensors while they slept. The result of the sleep test is an AHI (Apnoea-Hypopnea-Index) score, which is calculated by counting how many times the person stops breathing for more than 10 seconds at the time per hour.

Part of the data was used for training the model, while another smaller part of the data was kept out of the training to enable a subsequent test of the model’s accuracy. This test showed that the performance of the artificial intelligence was equal to the questionnaire-based method currently used for screening patients. In addition, the researchers asked a number of experienced sleep specialists to evaluate the 3D scans and predict each patient’s AHI score. Again, the data driven model’s performance was equivalent to the experienced doctors’ assessment.

More data and research needed

More data and research are needed before this method can be used in the clinic. PhD student Umaer Rashid Hanif says, “Originally the plan was to collect data from 30,000 patients. However, for different reasons the collection had to be concluded after only 1,500 patients. And this unfortunately limited the model’s performance. In my opinion, it would have been more accurate and better with the original amount of data. Soon, I will finish my PhD studies, but I hope that someone else will keep working with this area, as this disease has major consequences for the affected people. In addition, to feeling tired and not up to the mark during the day, sleep apnoea is connected with serious conditions such as cardiovascular diseases and dementia. Sleep has a huge impact on our health and wellbeing.”

Read more in the scientific paper published in IEEE Journal of Biomedical and Health Informatics

(Top photo: PhD Student Umaer Rashid Hanif tests his 3D scanning based screening method for sleep apnoea. Photo by Jesper Scheel)
21 JUNE 2024