Systematic review of automated sleep apnea detection based on physiological signal data using deep learning...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Sleep Medicine

Background:

• Sleep apnea (SLA) is a prevalent respiratory sleep disorder.
• Current diagnostic methods like polysomnography are costly and inconvenient.
• Automated detection systems using deep learning (DL) offer a potential solution.

Purpose of the Study:

• To comprehensively review and analyze recent deep learning applications for sleep apnea detection using physiological data.
• To classify and compare key characteristics of DL algorithms applied to pulse oximetry, ECG, airflow, and sound signals.
• To provide insights for researchers developing DL-based SLA detection systems.

Main Methods:

• Systematic literature review of 47 articles published between 2012 and 2022.
• Analysis focused on deep learning algorithms applied to 1-dimensional physiological data.
• Categorization based on physiological sensor data aspects and DL model perspectives.

Main Results:

• Deep learning demonstrates significant potential for processing physiological data (SpO2, ECG, airflow, sound) for SLA detection.
• Key factors influencing DL system performance include data input source, DL network architecture, training framework, and database.
• Studies were categorized by signal types, sampling frequency, window size, learning structure, and input data types.

Conclusions:

• Deep learning offers a promising avenue for developing automated, efficient, and accurate sleep apnea detection systems.
• Further research is needed to optimize DL models for extracting meaningful insights from 1D physiological signals.
• This review provides a detailed overview of current DL approaches, aiding future advancements in sleep apnea diagnostics.