Synthesis and comparison of analysing techniques for cardiac pathologies

Abstract

Phonocardiogram (PCG) signal processing helped uncover many crucial parameters that reflect the presence of a specific pathology. Therefore, we conducted a synthesis and comparison study focusing on research using various spectral and spectro-temporal techniques on PCG signals, such as the Fast Fourier Transform (FFT), Short Time Fourier Transform (STFT), Discrete Wavelet Transform (DWT), and Bispectral analysis. We extracted several parameters, including the frequency band, spectral entropy, entropy of approximation coefficients, temporal and frequency extents, entropy of phase, third-order moment, and more. We achieved promising results and high accuracies using these parameters for discriminating, classifying, and assessing the severity of cardiac pathologies. Finally, using complex feature-selection algorithms and machine learning/neural network classifiers, we compared the techniques and parameters to identify the optimal ones for cardiac severity assessment.