Time-frequency analysis of heart murmurs in children

Examination of heart sounds (phonocardiology) facilitates diagnosis of defects which manifest themselves as irregularities in the heart sound. Simply by listening to such sounds a trained physician can diagnose a variety of heart complaints, e.g. Atrial Septal Defects (ASDs) or colloquially “holes in the heart”. Such defects may not manifest themselves on ECGs. Any abnormality in a heart sound is generically termed a murmur. It is common practice for GPs to examine young children to check for the presence of a heart murmur. However, heart murmurs are relatively common in paediatric patients, the vast majority of these murmurs are nonpathological and the murmur will vanish, with no ill effects, over time. The problem arises in that GPs can detect the presence of a murmur, but cannot distinguish an “innocent” murmur from a pathological one. This classification task can be performed by specialist physicians simply listening to heart sounds. The goal of this work is to develop computer based methods capable of performing the task of the trained physician, i.e. the classification of heart murmurs. Heart sounds are clearly nonstationary signals and hence the natural analysis methods are those of time-frequency and/or time-scale. In this application there is no evidence to suggest that the analysis technique would benefit from a multiresolution type analysis, so the authors concentrate on time-frequency, rather than time-scale, methods. The study of heart sounds via time-frequency analysis has been undertaken by other authors