Transformer-based deep learning approach for obstructive sleep apnea detection using single-lead ECG.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Sleep Medicine

Background:

• Obstructive sleep apnea (OSA) is a common disorder caused by upper airway collapse during sleep, leading to significant health risks.
• Polysomnography (PSG) is the standard diagnostic tool but is expensive, time-consuming, and has long waiting lists.
• There is a growing need for accessible, accurate diagnostic tools for OSA, especially with advancements in machine learning (ML) and deep learning (DL).

Purpose of the Study:

• To introduce a novel transformer-based deep learning model for obstructive sleep apnea (OSA) detection.
• To evaluate the model's performance using a single-lead electrocardiogram (ECG) signal.
• To demonstrate an accurate and efficient alternative to traditional diagnostic methods like polysomnography (PSG).

Main Methods:

• Developed a transformer-based deep learning architecture capable of processing raw, high-sampling-rate ECG signals.
• The model was designed to handle high-noise data without preprocessing and preserve temporal continuity.
• Investigated various positional embedding techniques, including a novel autoencoder-based positional encoding method.

Main Results:

• The proposed DL model achieved a high F1 score, surpassing existing methods by over 13%.
• The model demonstrated robustness in handling raw, noisy ECG data.
• Achieved precise classification of apnea episodes at one-second intervals, offering detailed clinical insights.

Conclusions:

• A single-lead ECG-based deep learning model provides a highly accurate and efficient method for diagnosing obstructive sleep apnea (OSA).
• This approach offers a promising, less burdensome alternative to polysomnography (PSG) for OSA diagnosis.
• The model's ability to analyze raw signals and provide granular insights can significantly aid clinical decision-making.