Bilkent University
Department of Computer Engineering


Murmur Detection and Classification through Heart Sound Segmentation


Devrim ahin

Heart murmurs are pathologic heart sounds that originate from blood flowing through heart valves, due to physiological defects of the heart. Murmurs can be achieved via auscultation; that is, by listening with a stethoscope. However, precise manual detection and classification of murmur requires significant clinical expertise. Our objective is to develop a system which can help detection and classification of murmurs using heart sound recordings via 3M Littmann electronic stethoscopes.Although such classification algorithms exist; they depend on feature extraction from 'segmented' heart sound waveforms. Segmentation in this context refers to detecting and splitting cardiac cycles. However, heart sound signal is not a stationary signal; and has a low SNR ratio, which makes it very difficult to segment using the signal itself alone. Most of the commercial systems require an external ECG signal to determine S1 and S2 peaks, but ECG is not as widely available as stethoscopes. Segmentation through sound alone is not widely studied and there exists a few inefficient algorithms covering the topic. Moreover, no clinical ground truth for evaluation is available. We created a tool to enable manually segmentation of heart sounds, and construct a ground truth data set; on which we will evaluate and refine several approaches (including peak detection through envelograms, homomorphic filters and Hilbert transform) that we developed to achieve heart sound segmentation. Finally, we will train several machine learning algorithms to correctly classify murmurs.


DATE: 08 December, 2014, Monday @ 15:40