Design and performance evaluation of intelligent system to segregate and classify the phonocardiograph abnormalities using matched filter and multilayer perceptron-back propagation neural networks

Heart and cardiovascular diseases is considered as a major cause of increased mortality worldwide. Listening to the audio signal of the heart is effectively simple and non-invasive technique to detect the irregularities in heart function so that the correct analysis of it requires specialist knowledge and experience. The purpose of this paper is to design and implement an intelligent system for the isolation and classification of cardiac arrhythmias in the audio signal. We received heart sound signals through the heart sound recorder (PCG) in this system. The database were obtained from 41 informed volunteers and non-intervention form during a one month period from Cardiovascular Clinical Research Center Hospital in Sabzevar Vasei which included 15 patients with heart disease (10 males and 5 females with a mean age of 61±5. 7 years) and 26 healthy volunteers (19 males and 7 females with a mean age of 56. 5±8. 7 years) and finally the data were normalized. By testing the samples، 78 normal and 45 abnormal heart sound signals was selected from all subjects. By using adaptive filter، noise and environmental disturbances are extracted from the audio signals in pre-processing step of data. Feature extraction process will do based on cardiac cycles of sound signal in next step by applying the CWT transform on segmented signals and the 32-dimensions matrix of feature vectors will made up using wavelet coefficients component. The final classification of normal and abnormal heart sound signals is accomplished by using multi-layer perceptron neural network and back-propagation (MLP-BP). 70% and 30% of the data were used for training and testing the proposed neural network respectively based on our experiences. The proposed algorithm isolated original signal from noise signal and classified normal and abnormal sound signals at an acceptable level with an average accuracy of 96. 90%±1. 5% and 94%، respectively. This system has the potential for implementation in residential homes for people with different ages due to the reliability of the output of software in correct classification of audio signals and finally a person would be aware of his heart health.