Classification of wheeze in respiratory sounds by linear discriminant method

The aim of this study is classification of wheeze and non-wheeze using some selected features and detection of wheeze in respiratory sound signals acquired from patients with asthma and COPD. Taking into consideration that wheeze, having a sinusoidal waveform, has a different behavior in time-frequency domain from that of non-wheeze signals, features are chosen as kurtosis, Renyi entrophy, f₅₀/f₉₀ ratio and zero-crossing irregularity. Upon calculation of these features for each wheeze and non-wheeze portion, the whole data scattered as two classes in four dimensional feature space is projected using Fisher discriminant analysis (FDA) onto the single dimensional space that separates the two classes best. Observing that the two classes are visually well separated in this new space, Neyman-Pearson hypothesis testing is applied. Finally, the success rate has been found to be %94.9 for the training set, and leave-one-out approach together with the above methodology yields a success rate of %93.5 for the test set. Ending up with these results for both training and test sets, one can conclude that selected features and methodology are meaningful.