A Robust Approach for Pulmonary Disease Diagnosis with Multifractal Features of Lung Sounds Utilizing Machine Learning Models

Lung sounds (LS) contain crucial information about pulmonary diseases, but issues like ambient noise, motion artifacts and heart sounds can obscure important diagnostic features during auscultation, leading to misdiagnosis. Variational mode decomposition (VMD) is applied on raw LS for denoising. This study proposes an innovative approach to multifractal detrended fluctuation analysis (MFDFA) to extract multifractal features from the LS. The experimentation is conducted with recorded and online datasets, including 116 patients’ data from Thanjavur Medical College and Hospital (TMCH) and 118 patients’ data from a publicly available database. Six machine learning algorithms are implemented for intelligent pulmonary disease diagnosis. The results show that the quadratic discriminant analysis (QDA) classifier gives an F1 score of 96.33%, accuracy of 99.64%, precision of 96.18%, specificity of 99.65%, sensitivity of 99.54% and accuracy of 99.54%. This research highlights the robustness and superior performance of MFDFA-based features in classifying pulmonary diseases compared to existing methods.