Auditory Receptive Field Net Based Automatic Snore Detection for Wearable Devices.

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence

Background:

• Obstructive sleep apnea and hypopnea syndrome (OSAHS) diagnosis is often delayed due to inconvenient polysomnography (PSG) examinations.
• Snoring is an early indicator of OSAHS, making snore detection a valuable tool for early prediction.
• Wearable and IoT sensors offer potential for convenient, long-term sleep monitoring.

Purpose of the Study:

• To develop a deep learning-based model for accurate snore detection.
• To enable long-term home monitoring of snoring for early OSAHS prediction.
• To improve the discriminability of features between snoring and non-snoring sounds.

Main Methods:

• Proposed a deep learning model integrating an auditory receptive field (ARF) net for feature extraction.
• Implemented a detection model to predict snore events from sound waveforms using candidate boxes and confidence scores.
• Developed and utilized a snore detection dataset exceeding 4600 minutes for model evaluation.

Main Results:

• The proposed deep learning model demonstrated superior performance in snore detection.
• The model outperformed traditional approaches and existing deep learning models on the developed dataset.
• The auditory receptive field (ARF) net enhanced feature discriminability for improved accuracy.

Conclusions:

• The developed deep learning model offers an accurate and convenient method for long-term snore monitoring at home.
• This approach can facilitate earlier detection and prediction of obstructive sleep apnea and hypopnea syndrome (OSAHS).
• The integration of ARF nets shows promise for improving sound event detection in sleep studies.