Estimation of Respiratory Rate from Breathing Audio

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Digital Health

Background:

• The COVID-19 pandemic accelerated the adoption of remote healthcare, highlighting the need for non-contact vital sign monitoring.
• Accurate and inexpensive measurement of remote vital signs, particularly respiratory rate, remains a significant challenge in telehealth.
• Existing non-machine learning methods for respiratory rate estimation from audio have limited accuracy.

Purpose of the Study:

• To develop and evaluate a novel machine learning-based method for estimating patient respiratory rate using only audio signals.
• To address the limitations of existing methods and the scarcity of public datasets for respiratory rate estimation.
• To enable reliable and automated remote monitoring of respiratory rate for clinical applications.

Main Methods:

• A novel data augmentation technique was proposed to expand the effective size of a small, publicly available dataset, mitigating overfitting.
• The algorithm utilizes a spectrogram representation of audio signals and trains a recurrent neural network (RNN) to recognize breathing cycles.
• A data augmentation method was developed by exploiting the independence of periodic frequency components in spectrograms and permuting their order.

Main Results:

• The proposed machine learning method achieved a Mean Absolute Error (MAE) of 1.0 for respiratory rate estimation.
• The algorithm demonstrated a significant reduction in errors, nearly halving the errors of existing non-learning methods.
• The method relies solely on audio signals, which can be collected using standard smartphone devices.

Conclusions:

• Machine learning-based analysis of breathing sounds offers a promising approach for accurate remote respiratory rate estimation.
• This technology can enhance remote patient monitoring capabilities, supporting primary, specialty, and urgent care settings.
• The developed method provides a reliable and accessible tool for physicians to determine respiratory rate in remote patient evaluations.