Lung sound analysis for pulmonary diseases classification using machine learning algorithms

Abstract

Lung sound analysis has emerged as a promising non-invasive method for the early detection and diagnosis of pulmonary diseases. However, the presence of noise, such as ambient sounds, heartbeats, and motion artifacts, often distorts the lung sounds, making accurate diagnosis challenging. This study aims to address these challenges by proposing a novel approach for pulmonary disease classification through the analysis of lung sounds using machine learning algorithms. In this research, the lung sound signals are denoised using an Adaptive Variational Mode Decomposition (AVMD) technique. Additionally, a novel multifractal detrended fluctuation analysis (MFDFA)-based feature extraction method is proposed to enhance the analysis of lung sounds. Machine learning algorithms, specifically K-Nearest Neighbors (KNN) and Random Forest classifiers, are then employed to detect lung diseases. The study utilizes the publicly available ICBHI 2017 challenge database for analysis. Results indicate that the Random Forest classifier outperforms other models, achieving an accuracy of 99.10 %, precision of 96.15 %, and specificity of 98.75 %. These findings suggest that the combination of AVMD-based denoising, MFDFA-based feature extraction, and machine learning classification significantly enhances the performance of pulmonary disease diagnosis, offering a reliable and efficient tool for clinical applications.