Classification of respiratory sounds based on wavelet packet decomposition and learning vector quantization

Article type: Research Article

Authors: Pesu, L.^a | Helistö, P.^{a; *} | Ademovič, E.^b | Pesquet, J.-C.^b | Saarinen, A.^c | Sovijärvi, A.R.A.^c

Affiliations: [a] Laboratory of Biomedical Engineering, Helsinki University of Technology, FIN-02015, HUT, Finland | [b] Laboratoire des Signaux et Systèmes, ESE/CNRS and GDR/PRC ISIS, Université Paris-Sud, 91192 Gif sur Yvette, France | [c] Laboratory of Clinical Physiology, Helsinki University Central Hospital, Helsinki, Finland

Correspondence: [*] Correspondence to (current address): Pann Helistö, VTT Automation, Measurement Technology, P.O. Box 1304, FIN-02044 VTT, Finland. E-mail: [email protected].

Abstract: In this paper, a wavelet packet-based method is used for detection of abnormal respiratory sounds. The sound signal is divided into segments, and a feature vector for classification is formed using the results of the search for the best wavelet packet decomposition. The segments are classified as containing crackles, wheezes or normal lung sounds, using Learning Vector Quantization. The method is tested using a small set of real patient data which was also analysed by an expert observer. The preliminary results are promising, although not yet good enough for clinical use.

DOI: 10.3233/THC-1998-6108

Journal: Technology and Health Care, vol. 6, no. 1, pp. 65-74, 1998