Poster Abstract: A Comprehensive Approach for Cough Type Detection

Abstract

Presence of sputum in pulmonary system is an important bio-marker, critical in determining the existence of many disease such as lung infection, pneumonia, cancer, etc. While there has been many reports of successful algorithms to automatically detect cough instances, there has been not much work in identifying the cough type, or equivalently detection of sputum presence. Cough type detection is traditionally done by physical examination through hearing patients coughs in a clinical visit which is subjective and costly. This work tries to provide an objective comprehensive approach for cough type detection using an extensive set of acoustic features applied to the recorded audio from a relatively large population of both healthy subjects and patient with various pulmonary diseases and healthy controls. A total number of 5971 coughs (5242 dry and 729 wet) were collected from 131 subjects using Smartphone. Annotation was done using a crowd-source platform. Classification sensitivity and specificity values of 86% and 84% was achieved respectively which is the highest in literature to the best of our knowledge.