Separating obstructive and central respiratory events during sleep using breathing sounds: Utilizing transfer...

Area of Science:

• Biomedical Engineering
• Respiratory Medicine
• Artificial Intelligence in Healthcare

Background:

• Polysomnography (PSG) is the standard for sleep apnea diagnosis but is resource-intensive.
• Current portable devices often fail to distinguish between obstructive sleep apnea (OSA) and central sleep apnea (CSA).
• Accurate differentiation between OSA and CSA is crucial due to their distinct etiologies and treatments.

Purpose of the Study:

• To develop a non-invasive, cost-effective method for classifying obstructive and central sleep apnea events.
• To leverage tracheal breathing sounds and deep convolutional neural networks (CNNs) for apnea event differentiation.
• To provide a foundation for portable diagnostic devices capable of distinguishing between OSA and CSA.

Main Methods:

• Utilized transfer learning on six pre-trained CNNs (Alexnet, Resnet18, Resnet50, Densenet161, VGG16, VGG19).
• Fine-tuned CNNs using spectrograms of tracheal sound signals recorded during PSG from 50 participants.
• Trained and validated the model on a dataset containing both central and obstructive sleep apnea events.

Main Results:

• Achieved high accuracy in differentiating central from obstructive respiratory events.
• The combined CNN architecture demonstrated an overall accuracy of 83.66%.
• Reported sensitivity and specificity exceeding 83% for event classification.

Conclusions:

• Tracheal breathing sounds effectively distinguish between OSA and CSA.
• This methodology offers a less invasive and more accessible alternative to traditional PSG.
• The findings support the implementation of this technique in portable devices for enhanced sleep apnea diagnosis and personalized treatment.