Session S44.1

Heart Sound Extraction to Enable Remote Auscultation of Patients

N Mahmood*, T Syeda-Mahmood

Monta Vista High School
Cupertino, CA, USA

With the rising cost of healthcare of chronically ill patients, more and more healthcare providers are looking for solutions in which important patient data can be collected in a remote fashion while the patients remain at home. Auscultation, the act of listening for sounds made by internal organs through stethoscopes, is an early diagnostic procedure for abnormal sounds such as heart murmurs, and gallops, which will be useful to enable in a remote patient setting. Due to the inexperience of patients taking their own recordings, and due to ambient and conversational noise in the background, the signals received from remote auscultation devices are often noisy.
This paper addresses the problem of automatic segmentation of sound signals to identify heart-sound containing segments. The algorithm models the incoming sound signal as a short-time stationary signal consisting of various sounds occurring in successive time segments. An overlapping window analysis is performed and evidence of periodicity is checked within each overlapping segment using correlation of sound envelopes extracted from the regions. The sound envelopes are obtained using line approximations through a recursive partitioning algorithm. The peaks in the autocorrelation function correspond to the various periodicity patterns found in the signal. We note the most common inter-peak duration as representative of a heart beat duration.
A ground truth database of 513 original heart sound recordings taken from remote stethoscopes was manually segmented by physicians by listening to the sound signal and retaining heart sound containing regions. Of these recordings, the algorithm correctly identified heart sounds in 389 cases. The average overlap of time extent with manual identification was 73.4%. The missed cases mostly corresponded to arrhythmia cases and noise deeply intermixed with heart sound recordings.

(Abstract Control Number: 250)