Deep Learning-Based Event Counting for Apnea-Hypopnea Index Estimation Using Recursive Spiking Neural Networks.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Sleep Medicine

Background:

• Sleep apnea diagnosis relies on accurate event quantification, often requiring complex polysomnography.
• Current home screening methods face challenges in precise event localization, impacting reliability.
• The Apnea-Hypopnea Index (AHI) is a key metric for sleep apnea severity.

Purpose of the Study:

• To introduce RSN-Count, a novel Spiking Neural Network-based method for improved sleep apnea screening.
• To enable reliable estimation of the Apnea-Hypopnea Index (AHI) in home environments.
• To reduce the dependence on precise event localization for sleep apnea diagnostics.

Main Methods:

• Development of RSN-Count, a Spiking Neural Network technique for direct apneic event counting.
• Utilizing whole-night audio and SpO2 recordings for signal analysis.
• Focusing on event quantification rather than exact time-based pinpointing.

Main Results:

• RSN-Count demonstrated superior quantification of apneic events compared to established methods.
• Achieved lower Mean Absolute Error (MAE) in AHI estimation.
• Validated on a dataset of 33 participants with whole-night recordings.

Conclusions:

• RSN-Count offers a promising advancement for sleep apnea screening in home settings.
• The method enhances AHI estimation accuracy by focusing on event quantification.
• Addresses limitations in current diagnostics, potentially increasing accessibility and reducing reliance on polysomnography.