Differentiating Productive and Nonproductive Cough using Cough Acoustics & Machine Learning

Abstract

A wet cough by definition is when the cough brings about mucous in the phlegm due an irritant or an inflammation in the Lungs. Every cough regardless of it being wet or dry brings about phlegm and the degree of mucous varies and is much more in quantity in cases of a wet cough & can be used to determine the extent of SOB (shortness of breath), dyspnea due to excessive mucous. This is not always the case in that there are instances where the dry cough subjects may also experience dyspnea that can be attributed to other underlying & related conditions. From an acoustics perspective, a dry cough is usually something like a bland and hacking sound that is followed by an irritation in the Lungs while wet cough has a certain tonality to it whether it is crackling or a rattling sound. Tonality here can be measured using Spectral Flatness & other spectral variables that provide explainability are kurtosis, spectral centroid and spectral skewness. While this is not an exhaustive list, there are other variables that or of importance. Mucous is agnostic to whether it is a bacterial or viral infection but with Pneumonia, there are some indicators with sticky mucous with a bacterial infection Vs viral. However, an indicator of dyspnea induced by phlegm and whether it can differentiate between a bacterial and a viral Pneumonia is an important use case, the later in this context. Wet cough with wheezing caught at an early stage can determine an underlying Emphysema condition [1]. Emphysema: Types, Symptoms, Causes, Diagnosis, Treatment (health.com) or other forms of COPD such as chronic Bronchitis. Cystic Fibrosis has a certain type of mucous that if screened early can be priceless. The shortness of breath caused due to excessive mucus in the Phlegm, if it can rule out Lung Cancer as a screening modality are other areas where the applicability of wet cough can be useful. While the applications are many, in this paper we discuss the AI/ML methods on the acoustics of the “wetness or dryness” of a cough which provides an additional biomarker under the primary biomarker, “cough” in the field of acoustic epidemiology