A comprehensive respiratory evaluation: incorporating lung sound and disease classification along with spirometry assessment

Abstract

Respiratory disease, also known as pulmonary disease or lung diseases mainly affects the airways and hinders important functions of the lungs (NCI Dictionary of Cancer Terms). Some widely known respiratory diseases include asthma, pneumonia, Bronchiectasis, Bronchiolitis, chronic obstructive pulmonary disease (COPD), pulmonary fibrosis, upper respiratory tract infection (URTI), lower respiratory tract infection (LRTI), and lung cancer. Lung sounds are acoustic signals generated during breathing, commonly referred to as breath noises or respiratory sounds. They can offer insightful information about the condition of a patient’s lungs. Wheezing, crackles, or other abnormal lung noises can be a sign of underlying respiratory problems. On the other hand, procedures like Spirometry analyzes the volume and flow of air as a person breathes in and out to determine lung function. Spirometry may not always give a complete picture of a patient’s respiratory condition. This is where including lung sound analysis can be really helpful. Spirometry and lung sounds are both crucial instruments for evaluating respiratory health, but they have different roles and yield different kinds of data. While lung sounds provide qualitative details about the noises made when breathing, spirometry concentrates on quantitative measurements of lung function. In this paper, we explore ways in which we can make lung sound results more accurate and classifiable by using respiratory sound readings and by processing the data using machine learning and deep learning. We will be able to classify lung sound data into multiple categories. We will also be classifying spirometry data. In this research, we rigorously compare several machine learning and deep learning models to ascertain how well they classify lung sound and spirometry data. Gated Recurrent Unit (GRU), Support Vector Machine (SVM), Decision Tree, Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN) with different feature extractions, Stacked Autoencoder with SVM, and Attention and Vision Transformer are just a few of the models being examined. Through this assessment, we hope to find the best appropriate model(s) for improving the precision and usefulness of respiratory health evaluations, advancing the level of diagnostic capacities in the field of respiratory medicine.