Time and frequency-based approach to heart sound segmentation and classification

Abstract

In this study, we propose a decision tree classifier of heart sound signals. We determined repetitive fundamental heart sound segments based on adaptive similarity value clusterization of the sound signal, and we created a set of filters for decision tree parametrization. Using the filters together with inter-segment timings, we created three sets of markers: a set utilizing both S1 and S2 identification, a set where only one segment was identified, and a set without any identified segment. An individual classification tree was trained for each marker set. As a result, our classifier attained sensitivity (Se) of 0.66 and specificity (Sp) of 0.92 and overall score of 0.79 for a hidden random (revised) subset.