Abstract:
The heart is one of two organs which are crucial for human life. Therefore, a
disorder of the heart is of great importance to human health. Cardiovascular
diseases continue to be the leading cause of morbidity and mortality
worldwide. One of the first steps in evaluating the cardiovascular system after
detailed history taking is physical examination. Auscultation with a
stethoscope is regarded as one of the pioneer methods used in the diagnosis
of heart diseases. However, the fact that auscultation using a stethoscope
majorly depends on the skills of a physician‘s auscultation or his/her
experience, may lead to some problems in diagnosing heart problems. Due to
these difficulties, it is observed that the auscultation method has not been
very successful in determining heart diseases. Despite significant
interobserver variability, cardiac auscultation provides important initial clues in
patient evaluation and serves as a guide for further diagnostic testing.
Therefore, the use of an artificial intelligence method in the diagnosis of heart
sounds may help the physicians in a clinical environment.
In this study, an automatic technique is devised for the classification of
heart sounds. The study was carried on a total 120 heart sounds, and the
sounds were examined in two groups: normal heart sounds, and murmured
heart sounds. The study consisted of three stages. In the first stage the heart
sound signals are separated into sub-bands by using wavelet packet
transform.
In the second stage, which is the feature extraction stage, the entropy and
energy of each sub-band is calculated and then PCA, is used to reduce the
8
dimensionality of the feature vectors. The feature extraction stage is one of
the most significant stage affecting the performance of the classification.
In the third stage, the stage of classification is carried out using the KNN and
SVM. The two types of heart sound signals are used as an input of the K Nearest Neighbor and support vector machine and the sounds are classified
into two groups: normal sounds and murmur sounds. In the method used,
80% of the classification performance by KNN and 70% of the classification
performance by SVM is obtained
