DeepArousal-Net: A Multi-Block Recurrent Deep Learning Model for Proactive Forecasting of Non-Apneic Arousals From
Objective:
This study aimed to develop a deep learning model capable of accurately forecasting non-apneic sleep arousals, which are brief awakenings that disrupt sleep continuity and contribute to daytime fatigue.
Methods:
We introduce DeepArousal-Net, a novel deep learning model designed to predict non-apnea arousals using multichannel polysomnography (PSG) records. DeepArousal-Net employs a multi-block architecture composed of convolutional neural networks (CNNs) to extract features from a comprehensive set of PSG signals, including EEG, ECG, EOG, EMG, oxygen saturation, and airflow. Bidirectional Long Short-Term Memory (Bi-LSTM) layers are incorporated to capture temporal dependencies in the extracted features.
Results:
DeepArousal-Net achieved an accuracy of 81.31%, sensitivity of 71.23%, and specificity of 81.90% in forecasting arousals 30 seconds in advance. The model demonstrated superior performance compared to traditional time-series prediction methods.
Conclusion:
DeepArousal-Net's ability to forecast non-apneic sleep arousals marks a significant advancement over existing post-event detection systems.
Significance:
By anticipating arousals, DeepArousal-Net opens new possibilities for the development of innovative interventions and personalized sleep management strategies, potentially leading to improved sleep quality and overall well-being.
