Environment Knowledge-Driven Generic Models to Detect Coughs From Audio Recordings.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Respiratory Medicine

Background:

• Respiratory diseases, including COVID-19 and asthma, affect millions globally.
• Coughing is a common symptom requiring objective monitoring for diagnosis and care.

Purpose of the Study:

• To develop and evaluate environment-adaptive machine learning models for objective cough symptom reporting.
• To assess the utility of microphone sensing in aiding respiratory disease diagnosis.

Main Methods:

• Three machine learning approaches were developed: unguided, semi-guided, and guided.
• Models were trained and tested using multiple datasets, including open-sourced cough data.
• The approaches considered varying levels of user knowledge about the acoustic environment.

Main Results:

• Guided machine learning models demonstrated up to 28% higher accuracy compared to unguided models.
• The models showed applicable performance across diverse datasets, validating their generalizability.
• Accuracy improvements correlate with the degree of environmental knowledge incorporated into the models.

Conclusions:

• Guided machine learning models offer superior performance for cough analysis in respiratory disease detection.
• Effective implementation of guided models necessitates a comprehensive understanding of the surrounding acoustic environment.
• This approach holds potential for improved patient care through objective symptom assessment.