Classification of heart sounds using an artificial neural network

Multistage decision‐based heart sound delineation method for automated analysis of heart sounds and murmurs

“…Unlike other existing methods [14,19], the proposed method does not employ the searchback algorithm with sets of decision thresholds determined using the past detected segment durations and peaks for identifying the missed sound segments. Thus, the proposed method is quite straightforward and does not require any learning phase for accurately determining initial threshold parameters.…”

“…Heart murmurs are often characterised by the timing (early, mid, or late), intensity, duration, pitch (low, medium, or high), quality (blowing, rumbling, or musical), and configurations of crescendo, decrescendo, crescendo-decrescendo [1][2][3][4][5][6][7][8]. Thus, an automated delineation method for accurate measurements of sound parameters including amplitude, frequency content, duration, systolic, and diastolic intervals, timing, and configuration of murmurs is most important for effective diagnosis of cardiovascular diseases.Many heart sound segmentation (HSS) methods were reported based on the reference electrocardiogram (ECG) and/or carotid pulse (CP) signals [3], empirical mode decomposition (EMD) [4,5], hidden Markov models [7], wavelet transform (WT), and wavelet packet transform [8][9][10], temporal-spectral features [11][12][13][14], homomorphic envelogram, and self-organising probabilistic model [15], Hilbert transform [16,17], support vector machine [18], and artificial neural network [19]. On the basis of the feature extraction approaches, the HSS methods can be categorised into four major groups: (i) ECG and/or CP waveform-based methods, (ii) temporal-spectral feature-based methods, (iii) envelope-based methods, and (iv) hybrid methods.…”

“…Many heart sound segmentation (HSS) methods were reported based on the reference electrocardiogram (ECG) and/or carotid pulse (CP) signals [3], empirical mode decomposition (EMD) [4,5], hidden Markov models [7], wavelet transform (WT), and wavelet packet transform [8][9][10], temporal-spectral features [11][12][13][14], homomorphic envelogram, and self-organising probabilistic model [15], Hilbert transform [16,17], support vector machine [18], and artificial neural network [19]. On the basis of the feature extraction approaches, the HSS methods can be categorised into four major groups: (i) ECG and/or CP waveform-based methods, (ii) temporal-spectral feature-based methods, (iii) envelope-based methods, and (iv) hybrid methods.…”

Heart Sound Classification from Wavelet Decomposed Signal Using Morphological and Statistical Features

“…Wavelet coefficients (approximate and detail) were determined using Daubechies -2 wavelets for each cardiac cycle [14]. For each cardiac cycle, only the detail coefficients at a second decomposition level were used for further processing; this was based on the prior finding in the literature [6] that the detail coefficients at second decomposition level have the …”

“…Analysis of PCG signals, especially the automatic segmentation, and classification has been widely studied for past few decades. Tamer et al proposed a wavelet-based segmentation and artificial neural network based classification [6]. Guy et al proposed a clustering based approach in [7] [9,10].…”