Computer vision better diagnoses type of sleep apnea
August 12, 2020
TORONTO – At UHN’s KITE Research Institute, Scientist Dr. Babak Taati (pictured) and his research team have harnessed computer vision in a pioneering method for differentiating between different types of sleep apneas.
Sleep apnea is a chronic disorder in which breathing intermittently pauses during sleep. This interrupted breathing can dramatically increase the risk of heart disease, stroke and other complications.
Two types are used to categorize sleep apneas: obstructive, in which the throat temporarily collapses, blocking the airway; and central, in which the brain fails to send signals to the muscles that control breathing.
“Distinguishing sleep apneas as either obstructive or central is challenging but crucial to selecting an appropriate treatment,” says Dr. Taati. “This is because the treatments vary considerably depending on the type of sleep apnea – for example, continuous positive airway pressure therapy greatly benefits patients with obstructive sleep apneas, but is harmful for those with central sleep apneas.”
The new method developed by Dr. Taati’s team uses computer vision to monitor a sleeping patient and discern the type of apnea.
Videos of the patient sleeping are recorded with a night vision camera and then analyzed by a computer using artificial intelligence (AI) techniques. By tracking chest and abdomen movements of patients, the computer was able to learn the movements that corresponded to each apnea type.
To validate the method, the research team tested it on patients at KITE’s SleepdB Lab, headed by Scientist Dr. Azadeh Yadollahi. The team found that the computer could differentiate apneas with up to 95% accuracy.
As the first vision-based strategy for distinguishing sleep apneas, the new method does not require any monitoring equipment to be attached to the patient. Unlike other approaches, it does not disrupt the patient’s sleeping conditions.
This work was supported by FedDev Ontario, BresoTec Inc, the Natural Sciences and Engineering Research Council of Canada, the Toronto Rehabilitation Institute and Toronto Rehab Foundation.