Respiratory disease diagnosis using lung sounds

Lung sounds recorded from pathological and healthy subjects were classified as belonging to restrictive and obstructive respiratory diseases and healthy subjects. Feature parameters were obtained from autoregressive (AR) models applied to overlapping segments of respiratory sounds. Crackle parameters obtained from Prony model were further incorporated into the feature space for classification improvement. Two different multi-stage classifiers composed of k-nearest neighbor (k-NN) and voting and k-NN and multinominal classification were designed and their performances were compared.