Heart’s sound in several cases of hearts’ sick has special patterns which
can be recognized. Because of that heart’s sound is used to diagnose heart’s sick.
The technique which usually used is auscultation, hearing heart’s sound using
stethoscope. There are several problems with this technique, i.e. low frequency of
heart’s sound, low amplitude, noise factor, and likeness pattern between one types
of heart’s sound to the other type. To overcome these problems, it has been
developed a method heart’s sick detection using phonocardiogram analysis
(heart’s sound record) based on software.
This project aims to produce a tool to diagnose heart’s sound and classify
heart’s sick type, besides to analyze performance of orthogonal wavelet filter.
Generally, the system of heart’s sick detection consists of two main parts, i.e.
feature extraction using wavelet packet decomposition and feature classification
using Learning Vector Quantization (LVQ) neural network. Heart’s sound
spectral signal is divided using wavelet packet decomposition. Thus, Result of
decomposition process which several sub-band is calculated the energy to get
unique features. These features are recognized used LVQ neural network.
From experiment with feature extraction using wavelet filter coiflet 1 and
decomposition level 6 is obtained the accuracy of heart’s sick detection is 100%
for training data and 95,56% for testing data set.