Automatic cough detection via a multi-sensor smart garment using machine learning.

Area of Science:

• Biomedical Engineering
• Respiratory Medicine
• Machine Learning

Background:

• Cough quantification is crucial for managing conditions like asthma and COPD, but current methods (questionnaires, audio recordings) have limitations.
• Questionnaires lack accuracy due to recall bias, while audio recordings raise privacy concerns.
• Machine learning for cough detection often relies on microphones, posing similar privacy risks.

Purpose of the Study:

• To assess if a multi-sensor wearable device, excluding microphones, can accurately and unobtrusively detect coughs.
• To determine the relative importance of different sensor types (acceleration, respiration, electrocardiography) for cough detection.
• To explore a privacy-preserving alternative for objective cough quantification.

Main Methods:

• A multi-sensor smart-garment device measured 3D acceleration, respiration, and electrocardiography signals from 44 healthy adults.
• Participants performed various tasks, including coughs, breathing, talking, and laughing, in different postures.
• A Random Forest Classifier was trained and validated using inter-subject splits to predict cough events based on sensor data features.

Main Results:

• The dual-sensor model combining acceleration and respiration achieved the highest performance (F1-score 93.0%).
• Single sensors showed strong performance: acceleration (F1 92.6%), respiration (F1 88.9%), and electrocardiography (F1 77.5%).
• The system successfully distinguished coughs from other respiratory maneuvers, highlighting the value of acceleration and respiration data.

Conclusions:

• A multi-modal wearable device utilizing acceleration and respiration sensors can accurately detect coughs unobtrusively.
• This sensor-based approach offers a privacy-preserving alternative to audio recordings for cough quantification.
• Future research can leverage this technology for remote cough monitoring in clinical populations.