Efficient computation of image moments for robust cough detection using smartphones

Area of Science:

• Biomedical Engineering
• Digital Health
• Signal Processing

Background:

• Smartphone health monitoring apps show promise for improving quality of life and reducing healthcare costs.
• Existing applications struggle with respiratory disease monitoring due to inaccurate symptom measurement, particularly cough detection.
• Real-time cough detection on smartphones faces challenges with noisy audio signals and the need for computational efficiency to conserve battery life.

Purpose of the Study:

• To develop a robust and efficient smartphone-based system for real-time cough detection.
• To address the limitations of noisy input signals and high battery consumption in current cough monitoring technologies.
• To enable continuous and unobtrusive respiratory symptom tracking via mobile devices.

Main Methods:

• The system utilizes audio spectrograms and calculates local image moments.
• An optimized classifier is employed for accurate cough detection based on extracted features.
• The algorithm is designed for computational efficiency to minimize battery drain.

Main Results:

• The proposed system achieves high accuracy with 88.94% sensitivity and 98.64% specificity in noisy environments.
• It demonstrates significant computational efficiency, offering a 5500x speed-up compared to baseline implementations.
• Battery consumption is kept below 25% for 24-hour use (16% for the detector alone), representing a minimum 6x reduction compared to existing systems.

Conclusions:

• The developed smartphone cough detection system is robust, efficient, and suitable for real-world respiratory disease monitoring.
• The system effectively balances accuracy with low power consumption, addressing key limitations in current mobile health technology.
• This approach has the potential to significantly enhance patient quality of life and facilitate remote health management.