Segmentation and Parametrization of the Phonocardiogram for the Heart Conditions Classification in Newborns

Abstract

Phonocardiographs are analyzed for diagnostics of heart conditions in newborns. The algorithms of allocation of heart tones and selection of stationary periods on phonocardiograms are proposed. Dedicated heartbeats can be parameterized in different ways. The first set of parameters characterizes the shape and the time-amplitude features. The second set of parameters is the coefficients of the frequency-time decomposition of cardiac cycles with spline bases. This approach allows detecting of the Patent ductus arteriosus (PDA) by machine learning methods. Software for phonograms analysis has been developed.