UNIVERSITY PARK, Pa. — Patients suffering from sleep disorders could soon benefit from a new and faster way to diagnose and treat them, thanks to a new model proposed by an international team of researchers. The method utilizes deep learning to automatically classify a patient’s sleep stages.
“The International Classification of Sleep Disorders association has identified over 80 different sleep disorders with associated treatments,” said Guanjie Huang, a doctoral student in the Penn State College of Information Sciences and Technology who is leading the project. “A correct classification of the patient’s sleep stage is a prerequisite and an essential step to effectively diagnose and treat these sleep disorders.”
For the nearly 70 million Americans that suffer from one or more of these sleep disorders, such as insomnia, narcolepsy and apnea, their health could be negatively impacted if left untreated. The team’s proposed method could help doctors to speed up a diagnosis or allow patients to monitor sleep conditions on their own.
Huang explained that traditional sleep stage classification methods use polysomnography (PSG), a type of sleep study in which a sleeping patient wears up to several dozen sensors to collect physiological data, such as brain activity, eye movement and heart rhythm. Then, each segment of PSG data is either manually inspected by an expert or analyzed through traditional machine-learning methods.
Both of these methods are time consuming and require specific domain knowledge.
“In typical PSG methods, a patient needs to stay in a hospital and wear a bunch of sensors in order to collect the physiological data, while we only use data of a single channel electroencephalogram (EEG) sensor in our research,” said Huang. “It largely improves the user experience.”
After the data is collected, the proposed method utilizes a single, end-to-end, deep learning model to analyze the patient’s data and classify the sleep stage automatically.
“A faster and automatic classification of sleep stages can help patients monitor their sleep conditions by themselves,” said Huang. “Experts or doctors can also use it to speed up their inspection of sleep conditions or the diagnosis of a sleep disorder.”
“We used a more systematic approach, instead of trial and error, to determine and tune the deep learning architecture and parameters, which distinguishes our research from most deep learning studies,” added Chao-Hsien Chu, professor of IST at Penn State and a key faculty member of the research team. “Our experiments showed that the proposed deep learning-based method has better performance than previous work.”
The team, which also included Xiaodan Wu, professor of management science at Hebei University of Technology in Tianjin, China, earned the best paper award for their work, “A Deep Learning-Based Method for Sleep Stage Classification Using Physiological Signal,” at the 2018 International Conference on Smart Health, held July 1-3 in Wuhan, Hubei, China.
“[Receiving this award] indicates that a lot of people are interested in deep learning and healthcare, and that the deep learning method can outperform the traditional methods in some fields,” Chu concluded.