Overview
Researchers have made significant improvements in the ability of wearable health devices to accurately detect coughing, which can enhance the monitoring of chronic health conditions and help predict health risks, such as asthma attacks.
Key Findings
- Cough detection technologies have historically faced challenges in differentiating cough sounds from speech and other nonverbal noises.
- Edgar Lobaton, a professor at North Carolina State University, emphasizes that cough frequency is a crucial biomarker for tracking various respiratory conditions.
- Wearable health technologies can theoretically utilize machine learning to identify coughs, but real-world applications have proven more complex.
Challenges in Cough Detection
Despite advancements, models often struggle to differentiate coughs from:
- Human speech
- Sneezes
- Throat-clearing sounds
- Other similar noises
Lobaton explains that these models are trained on a library of sounds, but encounter difficulties when faced with unfamiliar noises.
Innovative Approach
To improve accuracy, researchers utilized data from wearable health monitors, specifically:
- Audio data captured by the monitors.
- Movement data from an accelerometer that detects coughing-related movements.
Yuhan Chen, the first author of the study, notes that while movement alone cannot identify coughing, combining sound and movement data enhances detection accuracy.
Results of the Study
In laboratory tests, the new model demonstrated:
- Increased accuracy compared to previous cough detection technologies.
- Fewer false positives, meaning sounds identified as coughs were more likely to be actual coughs.
Lobaton remarks that this advancement significantly improves the ability to distinguish coughs from both speech and nonverbal sounds, with ongoing efforts to further refine the technology.
Reference
Chen Y, Xiang F, Hernandez ML, Carpenter D, Bozkurt A, Lobaton E. Robust Multimodal Cough Detection with Optimized Out-of-Distribution Detection for Wearables. IEEE J Biomed Health Inform. 2025 Oct 2;PP. doi: 10.1109/JBHI.2025.3616945