Trend Statistics Network and Channel invariant EEG Network for sleep arousal study

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Sleep Medicine

Background:

• Sleep disorders significantly impact daily life and long-term health.
• Accurate sleep arousal scoring is crucial for diagnosing sleep-related breathing disorders.
• Current methods for sleep arousal detection can be labor-intensive and subjective.

Purpose of the Study:

• To develop an end-to-end trainable neural network for automated sleep arousal scoring.
• To improve the detection of respiratory effort related arousal (RERA) and estimation of respiratory disturbance index (RDI).
• To create a robust and objective method for analyzing sleep study data.

Main Methods:

• Proposed an end-to-end trainable neural network integrating a trend statistics network and a channel-invariant Electroencephalography (EEG) network.
• Utilized convolution networks and bi-directional long short-term memory (LSTM) for feature combination and arousal probability prediction.
• Developed an objective function optimizing for RERA and non-arousal regions, and a method for RDI estimation.

Main Results:

• Achieved a mean area under the precision-recall curve (AUPRC) of 0.50 for RERA detection in a 10-fold cross-validation on the Physionet Challenge 2018 database.
• Obtained a mean absolute error of 6.11 for RDI prediction.
• Demonstrated a two-class RDI severity prediction with 75% specificity and 83% sensitivity.

Conclusions:

• The proposed neural network offers an effective automated approach for sleep arousal scoring.
• The method shows promise in accurately identifying RERA and estimating RDI, aiding in sleep disorder diagnosis.
• This automated system has the potential to enhance the efficiency and objectivity of sleep analysis.