Heart sound segmentation based on recurrence time statistics

Heart sound segmentation is the primary step in automatic diagnosis of heart sounds. Since heart sound components have great diversity in frequency and amplitude, the focus of this paper is on time domain analysis. Time intervals between consequent peaks have been clustered in time domain and statistical data were extracted. Then a reference point was labeled by using the clustered data. We propose a novel algorithm to segment the heart sound signals, by using extracted data and the reference point. The performance of the algorithm has been evaluated using 240 periods of heart sound signals recorded from 12 subjects including normal and abnormal sounds. The algorithm has achieved a 93.8 percent precision and 100 percent of sensitivity during evaluation.