Heart Sounds Analysis and Classification Based on Long-Short Term Memory

Abstract

In this study, the development of an algorithm for the classification of heart sound phonocardiogram waveforms such as Normal, Murmur, Extrasystole, Artifact. By presenting the approach used for classification from a general machine learning application point of view, the types of classifiers used were detailed by comparing their features and their performance. The Long-Short Term Memory method which supports the classification of each cardiac cycle in sound recordings. In addition to the LSTM-based features, our method incorporates spectral features to summarize the characteristics of the entire sound recording.