Back
Thursday, 07 Sep, 2017
CTeD Features: Measuring the Zzzzs
It has been estimated that some 30% of the general population is affected by sleep disorders. And the figure is on the rise with the increased use of smartphones and tablets at night, regular consumption of caffeinated and sugary drinks, as well as climbing obesity levels. Sleep disorders are traditionally diagnosed by sleep experts who take various measurements of patients while they sleep, and then score the collected data to determine how well they slept. To improve this manual and subjective process, a team from Duke-NUS Medical School (Duke-NUS) has developed a new automated method for scoring sleep.
Potential applications of new sleep scoring process
The gold standard test for diagnosing sleep disorders is a polysomnography, also called a sleep study. The test monitors many body functions including brain waves, eye movements, heart rhythms, blood oxygen levels, and breathing during sleep. Once the test is completed, a trained specialist analyses the data for specific patterns. The sleep specialist typically scores the data in non-overlapping, 30 second sequential periods (or epochs), by assigning each epoch a sleep stage.
The process of conducting a sleep study is highly manual, time-consuming and costly. Furthermore, the scoring of the patient data can be subjective, with different sleep experts differing in their interpretation of a same set of data. To address some of these issues, Prof Mike Chee’s Lab at Duke-NUS has invented a system that can capture relevant patient measurements during a sleep study and automatically score the data to determine the patient’s sleep stages.
Mimicking the sleep expert using artificial intelligence (AI)
The team used an AI system to analyse a patient’s electroencephalography (EEG) and electrooculography (EOG) data, and score the patient’s sleep stages. The system is based on convolutional neural networks (CNNs), which are a type of artificial neural networks that are very effective in pattern recognition. CNNs are often used in the fields of image and video recognition as well as language processing.
The Duke-NUS’ system has been trained to recognise specific features in the patient’s EEG and EOG data, similar to how a sleep expert would identify such patterns during a sleep study. The system has been found to be robust, able to automatically score the patient data, and distinguish between different normal sleep stages. It is able to do this in real time, and more importantly, transform the highly manual traditional polysomnography process into one that is more cost-effective, rapid and reliable.
Duke-NUS applied for patents for this system in 2016. For more information on Duke-NUS IP related to Sleep Scoring, click here.