Classifying signals from a wearable accelerometer device to measure respiratory rate

Area of Science:

• Biomedical Engineering
• Medical Devices
• Physiological Monitoring

Background:

• Accurate respiratory rate measurement is crucial for hospitalized patients but challenging due to patient movement and breathing cycle variations.
• Traditional methods require prolonged observation periods (≥60 seconds) for adequate precision, which is often impractical in a clinical setting.

Purpose of the Study:

• To develop and validate an automatic system for measuring respiratory rate in acutely ill hospital patients.
• To assess the accuracy and reliability of a triaxial accelerometer-based system using machine learning for signal analysis.

Main Methods:

• A triaxial accelerometer attached to the chest wall measured breath duration in acutely ill patients.
• Machine learning, specifically a neural network, was trained to identify reliable respiratory signals from accelerometer data.
• The accelerometer's respiratory rate measurements were compared against a nasal cannula reference in a training and test set.

Main Results:

• Machine classification significantly reduced the median absolute difference in respiratory rate measurements to 0.48 breaths per minute, compared to 1.25 breaths per minute without classification.
• Despite rejecting 50% of recording periods as unreliable, the system provided accurate measurements even with only 10% of signal time classified as reliable.
• This approach offers greater reliability than manual nurse charting, which relies on much less observation time.

Conclusions:

• Chest wall-mounted accelerometers, when analyzed with machine learning, provide accurate respiratory rate measurements in hospital patients.
• This technology has the potential to enhance automatic illness scoring systems for adult inpatients.
• The system demonstrates improved accuracy and reliability for continuous respiratory monitoring in challenging hospital environments.