Effect of sensor type to respiratory sound classification

Abstract

Commonly known as “water in the lungs”, pulmonary edema is a respiratory condition caused by excess fluid build-up in air sacs of the lungs. It is often caused by congestive heart failure (CHF). As such, pulmonary edema is an indicative sign that a patient’s health condition is at risk, since CHF is not a condition to easily diagnose. The shortness of breath arising from pulmonary edema can lead to severe health risks if failed to be diagnosed early or left untreated. Although X-rays and computed tomography (CT) scans can detect water presence in the lungs, these methods are expensive and bulky, thus not easily made available to everyone. The aim of this project is to study and analyse the performances of three systems in classification of water presence in the lungs. This would be done using audio samples collected from using the systems on actual patients at Tan Tock Seng Hospital and from professional opinion of the patients’ health status. The analysis would be conducted based on 3 metrics – accuracy, sensitivity and specificity. From the analysis, both qualitative and quantitative results have shown that all three systems are relatively accurate with a minimum classification accuracy of around 80%. However, the superior system is System 2, where it has the most consistent performance with high levels of aforementioned metrics of around 98%.