Non-Contact Screening of OSAHS Using Multi-Feature Snore Segmentation and Deep Learning

Area of Science:

• Biomedical Engineering
• Sleep Medicine
• Artificial Intelligence

Background:

• Obstructive sleep apnea-hypopnea syndrome (OSAHS) is a common disorder linked to significant cardiovascular and metabolic risks.
• Previous research on snoring analysis for OSAHS often focused on detection or classification separately.

Purpose of the Study:

• To develop and evaluate a two-stage framework for integrated snoring event detection and OSAHS classification.
• To enhance automated screening for OSAHS using acoustic analysis of snoring.

Main Methods:

• An Adaptive Multi-Feature Fusion Endpoint Detection (AMFF-ED) algorithm was created for precise snore segmentation.
• A hybrid deep neural network, ERBG-Net (ECA-ResNet18 + Bi-GRU), was developed for snoring sound classification.
• Snore samples were converted to Mel spectrograms and processed by the deep learning model.

Main Results:

• The AMFF-ED algorithm achieved 96.4% accuracy in snore segmentation.
• The ERBG-Net model demonstrated 95.84% classification accuracy and a 94.82% F1 score on the test set.
• The framework successfully integrated detection and classification for OSAHS-related snoring.

Conclusions:

• The proposed two-stage framework provides accurate snoring detection and OSAHS classification.
• This approach shows promise for developing automated, at-home screening tools for obstructive sleep apnea-hypopnea syndrome.
• The integration of advanced signal processing and deep learning offers a novel method for sleep disorder analysis.