A deep learning-based algorithm for detection of cortical arousal during sleep.

Area of Science:

• Sleep Medicine
• Biomedical Engineering
• Artificial Intelligence

Background:

• Cortical arousals are key indicators of sleep quality and are used to identify hypopneic events.
• Electroencephalogram (EEG) recording for arousal detection is inconvenient for home sleep testing.
• Autonomic nervous system activity, observable via electrocardiography (ECG), often accompanies cortical arousals.

Purpose of the Study:

• To develop a deep learning algorithm for detecting sleep cortical arousals using only single-lead ECG data.
• To provide a more accessible and less invasive method for monitoring sleep quality during home sleep tests.

Main Methods:

• Utilized 1,547 polysomnography records from the Multi-Ethnic Study of Atherosclerosis database.
• Developed an end-to-end deep learning model combining convolutional and recurrent neural networks.
• The model processed raw ECG signals, extracted features, and determined arousal probability at 1-second resolution.

Main Results:

• The deep learning model achieved a gross area under the receiver operating characteristic curve of 0.93.
• A gross area under the precision-recall curve score of 0.62 was obtained on the test set (n=311).

Conclusions:

• The study demonstrates the potential of a single-lead ECG-based deep learning approach for accurate arousal detection.
• This method shows promise for improving the convenience and accuracy of home sleep testing.