English Abstract
Abstract :
This thesis provides a comparative study related to various signal processing and Artificial
Intelligence (AI) techniques applied to Phonocardiography signals (PCG) to automatically
differentiate normal heart sounds from five types of common murmurs. This should help to
provide a simple-to-use and cost-effective PCG-based system for the early detection of
cardiovascular anomalies. The average detection rate for heart anomalies in our study was
92.14%, while the classification rate was 71.02%. This is achieved by using Deep Neural
Network (DNN) classification with Hyperbolic Tangent (tanh) activation function, a 5-layer,
and 100 neurons per layer configuration. This thesis shows that the Discrete Wavelet
Transform (DWT) was found to be the best denoising algorithm and the Heart Sound
Envelogram (HSE) was the best segmentation method for the PCG signal. Mel Frequency
Cepstral Coefficients (MFCC) Features conjugated with DNN classification outperformed the
Time and Frequency-Domain counterparts. In the framework of smart and preventative health
care, the proposed system serves as a convenient home-care tool by providing potentially-ill
individuals with vital early warnings and guiding them to cardiologists for more precise
diagnoses
