A Deep Neural Network Model for Automated Heart Disease Classification

Heart related diseases presently pose one of the major threat worldwide. Heart abnormalities show a wide variation because of which accurate diagnosis becomes challenging. Phonocardigram (PCG) signals and their analysis has opened up a new paradigm in telemedicine. The abrupt fluctuations and the randomness of the PCG signals make them difficult to analyze and extract key parameters called features. Conventional Fourier techniques fail in this regard. In this paper, we have proposed a wavelet based technique wherein the discrete wavelet transform (DWT) have been used for the processing and feature extraction of the PCG signals has been done subsequently. The features extracted are energy, variance, entropy and standard deviation. The features extracted can be subsequently utilized for the classification of the PCG signals using the Conjugate Gradient Algorithm. The three categories of classified are: stenosis, regurgidation and normal. It has been shown that the proposed algorithm attains an accuracy of 93%.