Automatic cough episode detection using a vibroacoustic sensor.

Area of Science:

• Biomedical Engineering
• Respiratory Medicine
• Signal Processing

Background:

• Objective cough assessment is crucial for diagnosing respiratory diseases, as recommended by the European Respiratory Society.
• Automatic analysis methods are needed to quantify cough parameters accurately.
• Vibroacoustic signals from chest vibrations offer a potential data source for cough event detection.

Purpose of the Study:

• To evaluate the accuracy of artificial neural networks in classifying cough events using vibroacoustic signals.
• To explore the utility of chest-mounted sensors for objective cough monitoring.
• To compare different signal parametrization methods for cough event classification.

Main Methods:

• Participants (6 men, 8 women) produced imitated coughs, hand claps, speech, and shouting.
• Cough and other events were captured using chest vibration sensors and microphones.
• Feature vectors were extracted using time domain, time-frequency domain, and spectral modeling methods.
• Artificial neural networks were employed for automatic classification of the recorded events.

Main Results:

• Artificial neural networks achieved a classification accuracy of 95% for cough events.
• The vibroacoustic signals provided sufficient information for distinguishing coughs from other sounds.
• The applied parametrization methods were effective in feature extraction for the classification task.

Conclusions:

• Artificial neural networks effectively classify cough events using vibroacoustic signals with high accuracy.
• This non-invasive method shows promise for objective, quantitative cough monitoring in respiratory diagnostics.
• Further research can refine these techniques for clinical application in respiratory disease management.