Author ORCID Identifier

https://orcid.org/0000-0002-4369-2698

Document Type

Thesis

Date of Award

12-2021

Degree Name

Master of Science (MS)

Department

Computer Science

First Advisor

Dr. KC Santosh

Abstract

Significant changes have been made on audio-based technologies over years in several different fields along with healthcare industry. Analysis of Lung sounds is a potential source of noninvasive, quantitative information along with additional objective on the status of the pulmonary system. To do that medical professionals listen to sounds heard over the chest wall at different positions with a stethoscope which is known as auscultation and is important in diagnosing respiratory diseases. At times, possibility of inaccurate interpretation of respiratory sounds happens because of clinician’s lack of considerable expertise or sometimes trainees such as interns and residents misidentify respiratory sounds. We have built a tool to distinguish healthy respiratory sound from non-healthy ones that come from respiratory infection carrying patients. The audio clips were characterized using Linear Predictive Cepstral Coefficient (LPCC)-based features and the highest possible accuracy of 99.22% was obtained with a Multi-Layer Perceptron (MLP)- based classifier on the publicly available ICBHI17 respiratory sounds dataset [1] of size 6800+ clips. The system also outperformed established works in literature and other machine learning techniques. In future we will try to use larger dataset with other acoustic techniques along with deep learning-based approaches and try to identify the nature and severity of infection using respiratory sounds.

Subject Categories

Computer and Systems Architecture

Keywords

Respiratory, Sound Analysis, Lung Health

Number of Pages

66

Publisher

University of South Dakota

Share

COinS