Preliminary study in the analysis of the severity of cardiac pathologies using the higher-order spectra on the heart-beats signals

Abstract

Phonocardiography is a technique for recording and interpreting the mechanical activity of the heart. The recordings generated by such a technique are called phonocardiograms (PCG). The PCG signals are acoustic waves revealing a wealth of clinical information about cardiac health. They enable doctors to better understand heart sounds when presented visually. Hence, multiple approaches have been proposed to analyze heart sounds based on PCG recordings. Due to the complexity and the high nonlinear nature of these signals, a computer-aided technique based on higher-order statistics (HOS) is employed, it is known to be an important tool since it takes into account the non-linearity of the PCG signals. This method also known as the bispectrum technique, can provide significant information to enhance the diagnosis for an accurate and objective interpretation of heart condition. The objective expected by this paper is to test in a preliminary way the parameters which can make it possible to establish a discrimination between the various signals of different pathologies and to characterize the cardiac abnormalities. This preliminary study will be done on a reduced sample (nine signals) before applying it subsequently to a larger sample. This work examines the effectiveness of using the bispectrum technique in the analysis of the pathological severity of different PCG signals. The presented approach showed that HOS technique has a good potential for pathological discrimination of various PCG signals.

Bispectral analysis and information fusion technique for bearing fault classification

“…where X(f ) is the Fourier transform of x(t) and * denotes the conjugate complex. From equation ( 2) [15], it can be seen that the bispectrum is a function of two frequencies, and this characteristic allows it to capture and reveal the phase relationships between these frequencies in a signal. By analyzing the bispectrum, one can gain insights into how different frequency components within a signal interact and correlate in terms of phase.…”

“…So the bispectral values on the principal domain, i.e. triangular region, called the non-redundant region, Ω is calculated in this study [15,16]. This aim of the paper is to find the effectiveness of bispectrum in detecting bearing faults and not to find the phase relationships among the frequency components.…”

Heart Sounds Classification Based on High‐Order Spectrogram and Multi‐Convolutional Neural Network after a New Screening Strategy

“…The Algorithm 2 describes the learning process of the designed mCNN. All the convolutional layers with a kernel size of 3 × 3, in order to avoid overfitting, the BN layer is added after the convolutional layer, calculate the mean 𝜇 B and variance 𝜎 2 B for each batch of input y i , the 𝜇 B and 𝜎 2 B are used to standardize the input, the learnable parameters 𝛼 and 𝛽 are introduced to scale and shift the standardized xi as next layer input, behind the BN layer, an activation layer is used with the relu activation function, and all the maxpooling layers use the kernel of pool size 2 × 2. For binary or five classifications network models, the sudden drop from many neurons to few neurons in the last hidden layer will inevitably cause classification errors.…”

“…Moreover, the method is not enough for doctors to obtain information related to the mechanical activity of the heart information. [ 2 ] Cardiac auscultation based on heart sound recordings or phonocardiography remains the main screening tool for cardiac diseases. [ 3 ] At present, deep learning has achieved remarkable success in many practical classification tasks, exhibiting results that sometimes surpass those of humans.…”