Continuous Sound Collection Using Smartphones and Machine Learning to Measure Cough

Area of Science:

• Medical Technology
• Digital Health
• Respiratory Medicine

Background:

• Automated cough counting remains a challenge, hindering efficient evaluation of chronic cough, a significant unmet clinical need.
• Current methods for analyzing 24-hour cough frequency are often impractical for patient-centric, long-term monitoring.
• Developing automated solutions is crucial for advancing novel treatments for chronic cough.

Purpose of the Study:

• To propose and develop a smartphone-based system for continuous sound collection and automated cough detection.
• To enable objective, continuous assessment of cough frequency in a naturalistic setting.
• To lay the groundwork for incorporating patient-collected data into clinical trials for chronic cough.

Main Methods:

• Designed and verified a smartphone application for continuous sound collection using the phone's microphone.
• Developed a mathematical model for cough sound analysis and identification.
• Compared the performance of the cough detection model against commercial solutions and human raters.

Main Results:

• The smartphone application reliably performed continuous sound collection, with technical parameters verified for optimal configuration.
• The cough detection model achieved high sensitivity (90% at 99.5% specificity) and (75% at 99.9% specificity) on a public dataset.
• Initial experiments demonstrated promising results for applying cough detection techniques to patient-collected data.

Conclusions:

• The developed sound data collection application reliably collects and securely uploads audio data for analysis, serving as a vital first step for clinical trial integration.
• The initial cough detection model shows encouraging performance, paving the way for its use in future patient studies.
• This smartphone-based system has the potential to significantly ease cough symptom evaluation and support the development of new chronic cough treatments.