Feature Selection for Differential Diagnosis of Asthma and COPD: A Preliminary Study

Abstract

Asthma and chronic obstructive pulmonary disease (COPD) are two obstructive pulmonary diseases whose differential diagnosis is difficult due to overlapping symptoms and inadequacy of classical methods. The main aim of this study is to understand which combination of acoustic properties is more discriminative, and to develop a new method that can be used in clinical practice. Accordingly, a total of 26 features under eight different feature types have been calculated using pulmonary sounds acquired from 50 volunteers diagnosed with asthma (30 subjects) and 20 copd (20 subjects), and forward sequential feature selection has been performed using k-nearest neighbor (k-NN) classifier as it was observed to have better performance than Bayesian classifier on the singular features extracted. Consequently, the feature set composed of F-max, BIN5, AR(3), P-max, and F-95 has been selected to be the best feature set by the k-NN classifier with 96% accuracy in asthma and COPD discrimination. To develop a more reliable method for the differential diagnosis of asthma and COPD, the feature set should be augmented and different types of classifiers should also be used.