DIGITAL ANALYSIS OF CARDIAC ACOUSTIC SIGNALS IN CHILDREN

Sammanfattning: DIGITAL ANALYSIS OF CARDIAC ACOUSTIC SIGNALS IN CHILDREN

Milad El-Segaier, MD

Division of Paediatric Cardiology, Department of Paediatrics, Lund University Hospital, Lund, Sweden

SUMMARY

Despite tremendous development in cardiac imaging, use of the stethoscope and cardiac auscultation remains the primary diagnostic tool in evaluation of cardiac pathology. With the advent of miniaturized and powerful technology for data acquisition, display and digital signal processing, the possibilities for detecting cardiac pathology by signal analysis have increased. The objective of this study was to develop a simple, cost-effective diagnostic tool for analysis of cardiac acoustic signals. Heart sounds and murmurs were recorded in 360 children with a single-channel device and in 15 children with a multiple-channel device. Time intervals between acoustic signals were measured. Short-time Fourier transform (STFT) analysis was used to present the acoustic signals to a digital algorithm for detection of heart sounds, define systole and diastole and analyse the spectrum of a cardiac murmur. A statistical model for distinguishing physiological murmurs from pathological findings was developed using logistic regression analysis. The receiver operating characteristic (ROC) curve was used to evaluate the discriminating ability of the developed model. The sensitivities and specificities of the model were calculated at different cut-off points. Signal deconvolution using blind source separation (BSS) analysis was performed for separation of signals from different sources.

The first and second heart sounds (S1 and S2) were detected with high accuracy (100% for the S1 and 97% for the S2) independently of heart rates and presence of a murmur. The systole and diastole were defined, but only systolic murmur was analysed in this work. The developed statistical model showed excellent prediction ability (area under the curve, AUC = 0.995) in distinguishing a physiological murmur from a pathological one with high sensitivity and specificity (98%). In further analyses deconvolution of the signals was successfully performed using blind separation analysis. This yielded two spatially independent sources, heart sounds (S1 and S2) in one component, and a murmur in another.

The study supports the view that a cost-effective diagnostic device would be useful in primary health care. It would diminish the need for referring children with cardiac murmur to cardiac specialists and the load on the health care system. Likewise, it would help to minimize the psychological stress experienced by the children and their parents at an early stage of the medical care.