Optimization of a Deep-Learning-Based Cough Detector Using eXplainable Artificial Intelligence for Implementation on Mobile Devices

Abstract

Respiratory diseases, including COPD and cancer, are among the leading causes of mortality worldwide, often resulting in prolonged dependency and impairment. Telemedicine offers immense potential for managing respiratory diseases, but its effectiveness is hindered by the lack of reliable objective measures for symptoms. Recent advances in deep learning have significantly enhanced the detection and analysis of coughing episodes, a key symptom of respiratory conditions, by leveraging audio signals and pattern recognition techniques. This paper introduces an efficient cough detection system tailored for real-time monitoring on low-end computational devices, such as smartphones. By integrating Explainable Artificial Intelligence (XAI), we identify salient regions in audio spectrograms that are crucial for cough detection, enabling the design of an optimized Convolutional Neural Network (CNN). The optimized CNN maintains high detection performance while significantly reducing computation time and memory usage.