Design For Real Time Heart Sounds Recognition System

ABSTRACT

Auscultation is a technique, in which Physicians used the stethoscope to

listen to patient’s heart sounds in order to make a diagnosis. However, the

determination of heart conditions by heart auscultation is a difficult task and

it requires special training of medical staff. On the other hand, in primary or

home health care, when deciding who requires special care, auscultation

plays a very important role; and for these situations, an ‘‘intelligent

stethoscope’’ with decision support abilities is highly needed and it would

be a great added value.

In this study a reliable Real Time Heart sounds recognition system has been,

introduced, designed, implemented and successfully tested.

The system algorithm has been realized in two phases, offline data phase and

real data phase. For offline data phase, 30 cases of Heart Sounds (HSs) files

were collected from medical students and doctor's world website, and then

the background noise is minimized using wavelet transform. After that,

graphical and statistics features vector elements are formed for both time and

frequency domain. Finally, classification process was accomplished using

look-up table. The implementation of the proposed algorithm produced

accuracy of 90%, and sensitivity of 87.5%.

In experimental phase (real time data), electronic stethoscope has been

designed and recorded HSs directly from 30 volunteers with 17 normal case

and 13 various pathologies cases. In preprocessing stage, an adaptive filter

was used to filter heart sounds from lung sounds, due to lung sound

overlapped with heart sound in sub frequency band. Then, wavelet was

applied to minimized background noise and features are formed for

classification process, as well as offline data phase. The implementation of

the proposed algorithm produced accuracy of 80%, and sensitivity of 82.4%.

The advanced steps for implementing a portable module by embedded DSP

have been successfully achieved. Firstly, System SIMULINK model was

built, and then real time workshop was used to generate embedded coder,

finally the code files linked to Code Composer Studio Software and running the project successfully.