Unsupervised Clustering of Heart Sound Recordings for Cardiac Auscultation Database Indexing

This study proposes an unsupervised framework for classifying heart sound data. Its goal is to cluster unknown heart sound recordings, such that each cluster contains sound recordings belonging to the same heart diseases or normal heart beat category. The proposed framework is more flexible than the conventional supervised classification of heart sounds by the case when heart sound data belong to undefined categories or when there is no prior template data for building a heart sound classifier. The proposed system includes four components, heart sound feature extraction, similarity computation, cluster generation, and estimation of the optimal number of clusters. Our experiments show that the resulting clusters based on our system are roughly consistent with the heart beat categories defined by human labeling, which indicates the feasibility of the unsupervised classification framework.