Feasibility of Cough Detection and Classification Using Artificial Intelligence in an Ambulatory Setting with a Ceiling Mounted Microphone

Cough is a sign of numerous respiratory infections and is often quantified by cough frequency. Although the need for accurate and objective cough detection in ambulatory settings is widely acknowledged in the medical literature, little research has been done on automating the classification using a single microphone in an open, real-world setting. This study examined the feasibility of applying artificial intelligence to recognize and categorize coughs by patients wearing or not wearing masks in a waiting room of a primary care institution with a single microphone and varying degrees of background noise. A sequential convolutional neural network (CNN) consisting of two 2D convolutional layers with 3x3 kernels and four filters were used with varying parameters. The best performing classification model used three layers with 64, 32 and 16 filters. It achieved an overall accuracy of 98.5% with a sensitivity of 98.2% and specificity of 98.8%. The findings imply that detection using artificial intelligence and a single microphone in a waiting room might be feasible to use in certain scenarios.