From: Localization and classification of heart beats in phonocardiography signals —a comprehensive review
Study | Year | Pre-processing techniques | Localization techniques | Classification techniques | Beat types | Data set | Signals statistics | Evaluation metrics | Results |
---|---|---|---|---|---|---|---|---|---|
Jusak et al. [98] | 2016 | CEEMD and EEMD | Pearson distance metric | Pearson distance metric | Murmur | University of Michigan Heart Sound and Murmur Library [26] | 10 signals | ΔSNR | – |
Salman et al. [97] | 2016 | EMD denoising | – | – | Synthetic normal and abnormal heart sound corrupted by noise | University of Michigan Heart Sound and Murmur Library [26] | – | Signal-to-noise ratio (SNR), root mean square error (RMSE), and percent root mean square difference (PRD) | – |
Banerjee et al. [95] | 2016 | Median and low pass filtering | VMD, normalization, Shannon energy, and dynamic thresholding | VMD, normalization, Shannon energy, and dynamic thresholding | Normal and abnormal heart sounds | PASCAL CSHC 2011 dataset A and B | 10 signals | Average detection time error | – |
Salman et al. [83] | 2015 | Signal decimation, low pass filtering, and EMD denoising | Time-based peak detection of Shannon energy envelope, and Heron’s formula | Time duration between peaks | Normal heart sounds | University of Michigan Heart Sound and Murmur Library | 8 signals of 70 s | – | – |
Salman et al. [101] | 2015 | Wavelet transform, total variation, and EMD | Time-based peak detection of Shannon energy envelope and Heron’s formula | Wavelet transform, total variation, and EMD | Normal heart sounds, S3, and S4 gallop | University of Michigan Heart Sound and Murmur library | – | – | – |
Papadaniil et al. [92] | 2014 | Median and low pass filtering | Energy, instantaneous frequency, bootstrap, and kurtosis-based criterion | Kurtosis-based S1 and S2 detection | Normal heart sounds, aortic stenosis, and mitral regurgitation | First Cardiac Clinic of Papanikolaou General Hospital in Thessaloniki, Greece | 43 signals composed of 2602 cycles | Mean prediction Power and mean accuracy | 74–100% and 45–100% |
Varghees et al. [100] | 2014 | Low pass filtering and total variational filtering | Shannon entropy envelope computation, instantaneous phase-based boundary determination, and boundary location adjustment | Shannon entropy envelope computation, instantaneous phase-based boundary determination, and boundary location adjustment | Normal heart sounds, fixed and wide split, early systolic, late systolic and pan systolic, mitral and tricuspid stenosis, aortic and pulmonic regurgitation, mitral prolapse, ejection murmur and clicks, and different time-varying systolic and diastolic murmurs | E-General Medical | – | Average sensitivity and positive predictivity | 99.43 and 93.56% |
Jimenez et al. [94] | 2014 | Resampling, segmentation, and normalization | EMD, EEMD EMD with adaptive noise | Features based on ergodic HMM, MFCC, and statistical moments of HHT | Normal heart sounds and murmurs | Private dataset | 400 beats | Accuracy, specificity, and sensitivity | For all measures above 95 |
Gavrovska et al. [88] | 2013 | Wavelet-based denoising and lifting scheme | DWT, EMD, and EEMD | DWT, EMD, and EEMD | Normal heart sounds and clicks | – | – | Comparison in terms of SNR and ratio R | – |
Boutana et al. [91] | 2013 | – | EMD | EMD | Early aortic stenosis, late aortic stenosis, mitral, and aortic regurgitation | E-General Medical | 4 signals | – | – |
Sun et al. [90] | 2013 | EMD followed by wavelet decomposition | Shannon energy envelop from denoised signal as well as from EMD channels of original signal | Shannon energy envelop from denoised signal as well as from EMD channels of original signal | Normal heart sounds | – | 30 signals | Accuracy | 99.74% |
Zhao et al. [85] | 2013 | Denoising and framing | Hilbert-Huang transform, normalization, and DCT | VQ with LBG algorithm | Normal Heart sounds | 40 participants and 7 heart sounds for each collected by the University of Catania, Italy | 280 samples of 10 s each | Correct recognition rate | 94.16% in marginal spectrum system and 84.93% in Fourier spectrum |
Moukadem et al. [89] | 2012 | Decimation and high pass filtering | S-transform and radial basis functions (RBF) neural network | Features based on SVD performed on S-transform, EMD, and KNN | Normal heart sounds | Data collected at Hospital of Strasbourg, France | 80 signals of 8–12 s each | Sensitivity and specificity | 94 and 97% |
Segalen et al. [93] | 2011 | – | Modulated-EMD | Modulated -EMD | Normal Heart sound | Private dataset | 20 signals | – | – |
Zhao Zhidong [84] | 2005 | – | Instantaneous frequency of diastolic murmurs | Instantaneous frequency of diastolic murmurs and SVM | Normal heart sounds and coronary artery disease | Private dataset | 77 signals | Correct and incorrect diagnosis | 17 out 20 for abnormal and all normal correctly diagnosed |
Sun et al. [87] | 2005 | – | Hilbert transform on IMF given by EMD | Hilbert transform on IMF given by EMD | Normal and abnormal heart sounds | Private dataset | 2 signals | – | – |
Sun et al. [86] | 2004 | – | EMD | EMD | Normal and abnormal heart sounds | Private dataset | 2 signals | – | – |