A RR interval based automated apnea detection approach using residual network.

Area of Science:

• Cardiology
• Biomedical Engineering
• Sleep Medicine

Background:

• Sleep-disorder breathing, particularly apnea, affects a growing global patient population.
• Current polysomnography (PSG) diagnosis is complex and time-consuming, leading to underdiagnosis.
• There is a need for precise, accessible diagnostic support systems using electrocardiograph (ECG).

Purpose of the Study:

• To develop an effective and precise diagnosis support system for apnea detection using ECG.
• To propose a novel approach utilizing residual networks and RR intervals for apnea detection.
• To investigate the potential of representing RR intervals using neural networks.

Main Methods:

• Application of a residual network to analyze information from RR intervals.
• Introduction of a dynamic autoregressive representation for interpreting RR intervals via convolutional layers.
• Training and testing the model on a publicly available Physionet dataset (30 recordings for training, 5 for testing).

Main Results:

• Achieved high performance in per-segment apnea detection: 94.4% accuracy, 93.0% sensitivity, and 94.9% specificity.
• Outperformed prevalent RR interval-based methods and deep neural networks using original ECG signals.
• Demonstrated good adaptivity with ECG-derived respiration (EDR) signals and required fewer input samples than other deep learning models.

Conclusions:

• A deep residual network effectively detects apnea using low-sample-rate RR intervals.
• The study validates the potential of neural networks for representing RR interval data.
• The model exhibits strong adaptability, particularly when utilizing EDR input.