Fourier Decomposition-Based Automated Classification of Healthy, COPD, and Asthma Using Single-Channel Lung Sounds

Abstract

Subjective discrimination of asthma and Chronic Obstructive Pulmonary Disease (COPD) is challenging as they share overlapping symptoms and are subject to personal interpretation. Hence, there is a demand for an alternative diagnostic system devoid of any subjective interference. The current study introduces Fourier Decomposition Method (FDM) based models utilizing Discrete Cosine Transform (DCT) and Discrete Fourier Transform (DFT) to identify patients with asthma and COPD by analyzing lung sound signals. The signals were decomposed into Fourier intrinsic band functions (FIBF) using three filter banks: dyadic, equal energy, and uniform band. Four statistical attributes, namely: Shannon entropy, log entropy, median absolute deviation and kurtosis, are calculated from relevant FIBF. Support vector machine (SVM), k-nearest neighbor (kNN) and ensemble classifier (EC) optimized with Bayesian optimization are used for classification of asthma vs COPD and normal vs adventitious sound, respectively. The highest accuracies achieved using DCT with 10-fold cross-validation are as follows: 99.4% (Asthma vs COPD), 99.1% (Asthma vs COPD vs Normal), 99.4% (COPD vs Normal) and 99.7% (Asthma vs Normal). Similarly, the highest accuracies reported by DFT with 10-fold cross-validation are: 99.4% (Asthma vs COPD), 99.6% (Asthma vs COPD vs Normal), 99.4% (COPD vs Normal) and 99.8% (Asthma vs Normal).