Pcg Event Detection and Classification Using Right Singular Vectors

Abstract

An unsupervised methodology for detecting and clas- sifying recurring sounds in the phonocardiogram (PCG) without reference to any other biological signal is pro- posed. Peaks in the orthonormal time domain singular vectors of a time-frequency representation of the PCG are used as both detection statistics and classification features. Applying the proposed strategy to the first three singular vectors detects 97.5% of the 1224 fundamental heart sounds in 42 PCG’s.