Automatic snoring detection using a hybrid 1D-2D convolutional neural network.

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence

Background:

• Snoring significantly impacts the quality of life for individuals with sleep-disordered breathing and their partners.
• Snoring serves as a critical indicator for screening and diagnosing obstructive sleep apnea (OSA).
• Accurate detection of snoring from nocturnal respiratory audio is essential but challenging.

Purpose of the Study:

• To develop a high-precision, automatic snoring detection algorithm.
• To design a non-contact system for recording sleep respiratory audio in a natural environment.
• To evaluate a hybrid convolutional neural network (CNN) model for snore detection.

Main Methods:

• Utilized a non-contact data acquisition system to record nocturnal sleep respiratory audio.
• Developed a hybrid CNN model incorporating 1D CNN for raw signal processing and 2D CNN for visibility graph representations.
• Employed a hybrid CNN architecture for automatic snore detection from audio recordings.

Main Results:

• The proposed hybrid CNN model achieved an average classification accuracy of 89.3%.
• The algorithm demonstrated high performance with an average sensitivity of 89.7% and specificity of 88.5%.
• An average Area Under the Curve (AUC) of 0.947 indicates robust detection capabilities, outperforming existing models.

Conclusions:

• The developed method is effective and significant for large-scale screening of obstructive sleep apnea (OSA) patients.
• This approach offers a valuable alternative framework for time series analysis in sleep research.
• The non-contact, automated system facilitates OSA screening in everyday settings.