ResSpec: Classification of Respiratory Sounds Using Machine Learning Algorithms

Abstract

This study develops a respiratory sound classification system to assist respiratory therapists in identifying and classifying respiratory sounds using machine learning. ResSpec categorizes sounds like crackles and wheezes, which are indicative of conditions such as asthma and pneumonia. Using annotated datasets, it supports clinical decision-making and facilitates the early identification of respiratory issues. Evaluation through the System Usability Scale (SUS) survey, involving licensed respiratory therapists, yielded an SUS score of 82.91, indicating "excellent" usability. The system proved effective for clinical support and educational purposes, particularly in detecting crackles and wheezes with reasonable accuracy. While promising, limitations include the need for a larger dataset to improve classification of other respiratory sounds. Future efforts will focus on expanding the dataset, optimizing the model, and addressing current limitations to broaden its utility in healthcare.