Classification of Breathing Phase and Path with In-Ear Microphones.

Area of Science:

• Biomedical Engineering
• Machine Learning Applications
• Wearable Health Technology

Background:

• Smart in-ear devices (hearables) are increasingly used for health monitoring.
• Existing in-ear breath monitoring relies on peak detection, which is sensitive to movement artifacts.
• There is a need for robust methods for continuous breath monitoring using hearables.

Purpose of the Study:

• To classify breathing path and phase using an in-ear microphone.
• To evaluate the effectiveness of machine learning, specifically XGBoost, for this classification task.
• To demonstrate the feasibility of hearables for continuous breath monitoring.

Main Methods:

• Utilized an existing database of breathing sounds captured by an in-ear microphone.
• Employed XGBoost, a machine learning classifier, due to the small dataset size.
• Trained and evaluated classifiers for binary path, binary phase, and four-class path and phase identification.

Main Results:

• Achieved 86.8% accuracy for binary path classification.
• Obtained 74.1% accuracy for binary phase classification.
• Reached 67.2% accuracy for four-class path and phase classification, outperforming existing algorithms in recall and F1 score.

Conclusions:

• The study demonstrates the feasibility of using hearables for continuous breath monitoring.
• XGBoost provides a reliable machine learning approach for classifying breathing patterns from in-ear recordings.
• This technology has the potential to enhance remote health monitoring capabilities.