Wireless Earphone-based Real-Time Monitoring of Breathing Exercises: A Deep Learning Approach.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Deep breathing exercises are crucial for many therapies, but monitoring patient compliance at home is challenging due to a lack of specialized equipment.
• Existing methods for therapy compliance monitoring are often limited to clinical settings, hindering consistent patient engagement.

Purpose of the Study:

• To develop and validate a framework for accurately monitoring at-home breathing exercises using consumer-grade hardware.
• To assess patient compliance with breathing therapies in a home environment.

Main Methods:

• A novel framework utilizing wireless earphones and smartphones for real-time breathing monitoring.
• Two convolutional neural networks (CNNs) were employed: a channel classifier (nasal, oral, pause) and a phase classifier (inhalation, exhalation).
• Processing of 500 ms audio signals for accurate breathing phase and channel detection.

Main Results:

• The channel classifier achieved a maximum F1 score of 97.99%, distinguishing nasal, oral breathing, and pauses.
• The phase classifier achieved a maximum F1 score of 89.46% in determining inhalation and exhalation.
• Demonstrated high accuracy in real-time breathing channel and phase detection using commodity earphones.

Conclusions:

• Commodity earphones can be effectively utilized for real-time breathing channel and phase detection.
• The proposed system shows significant potential for facilitating breathing therapy compliance monitoring in at-home settings.
• This technology can enhance the management and outcomes of therapies requiring regular breathing exercises.