|
| Year | Related work | Dataset | Classes | Feature extraction | Algorithm | Accuracy (%) |
|
| 2021 | Singh et al. [17] | PhysioNet 2016 | 2 | Wavelet decomposition, homomorphic filtering, Hilbert transform, and power spectral density without segmentation | AlexNet | 90.00 |
| 2020 | Krishnan et al. [15] | PhysioNet 2016 | 2 | Feedforward neural network | Feedforward neural network | 85.65 |
| 2020 | Hu et al. [18] | PhysioNet 2016 | 2 | 1D-CNN | ANN | 94.60 |
| 2021 | Zeng et al. [19] | PhysioNet 2016 | 2 | Tunable Q-factor wavelet transform, variational mode decomposition, phase space reconstruction | Radial basis neural network | 97.89 |
| 2021 | Tuncer et al. [20] | Yaseen dataset | 5 | Petersen graph pattern | KNN, DT, LD, BT, SVM | KNN: 100.00
DT: 95.10
LD: 98.30
BT: 98.60
SVM: 99.90 |
| 2018 | Yaseen et al. [21] | Yaseen dataset | 5 | MFCC and DWT | SVM, KNN, and DNN | SVM: 97.90
KNN: 97.40
DNN: 92.10 |
| 2020 | Chen et al. [22] | PhysioNet 2016 | 2 | Modified frequency slice wavelet transform | CNN | 94.00 |
| 2016 | Abdollahpur et al. [23] | PhysioNet 2016 | 2 | Shannon entropy | 5 groups of features from the time-domain, time-frequency domain, and perceptual features such as Shannon entropy, MFCC, etc | 92.48 |
| 2020 | Ghosh et al. [24] | Yaseen dataset | 4 | Chirplet transform, local energy, and local entropy | Sparse representation classifier | 98.54 |
| 2020 | Ghosh et al. [25] | Yaseen dataset | 5 | Spline kernel-based chirplet transform, L1-norm, sample entropy, and permutation entropy | Deep layer kernel sparse representation network | 95.67 |
| 2019 | Ghosh et al. [26] | Yaseen dataset | 4 | Statistical features extracted from time-frequency magnitude and phase matrix of segmented PCG signals | Random forest | 93.91 |
| 2022 | Karhade et al. [27] | Yaseen dataset and PhysioNet 2016 | 4 | Time-frequency images obtained using both time domain polynomial chirplet transform (TDPCT) and frequency-domain polynomial chirplet transform (FDPCT) | Deep convolutional neural network | TF images obtained from TDPCT: 99.00
TF images obtained from FDPCT: 99.48 |
| 2019 | Singh and Majumder [28] | PhysioNet 2016 | 2 | Wavelet decomposition, Hilbert transform, homomorphic filtering, and power spectral density | KNN | 90 |
| 2020 | Yang et al. [29] | PASCAL and PhysioNet 2016 | 2 | Time domain features based on envelope extracted signal processed by EMD | SVM | 96.67 |
| 2021 | Tseng et al. [30] | PhysioNet 2016 | 2 | — | Booster LKNet | 92.48 |
| 2021 | Alkhodari and Fraiwan [31] | Yaseen dataset | 5 | — | CNN-BiLSTM | 99.32 |
| 2021 | Sankararaman [32] | 48 mitral incompetence and healthy heart sound signals | 2 | Wavelet scalogram generated by CWT | KNN | 100 |
| 2018 | Latif et al. [33] | PhysioNet 2016 | 2 | MFCC from 25 ms of the window with a step size of 10 ms | RNN with BiLSTM units | 98.61 |
| 2017 | Zhang and Wei [34] | PhysioNet 2016 | 2 | 20 waveform features extracted from segmented signals and 15 power spectral density features extracted from several frequency ranges | Least squares support vector machine (LS-SVM) | 86.85 |
| 2022 | Morshed et al. [35] | PhysioNet 2016 and Yaseen dataset | 2/5 | Magnitude, frequency, and phase of each Burg’s spectrum along with statistical features | Ensembled bagged trees | Yaseen dataset: 99.28
PhysioNet 2016: 93.46 |
| 2022 | Tariq et al. [36] | PASCAL dataset | 6 | Spectrogram and chromagram | FDC-FS network | 97.00 |
| 2022 | Sun et al. [37] | 665 AR, 381 AS, 315 ASD, 769 MR, 439 MS, 1056 NM and 327 VSD sounds | 7 | Frequency features extracted from envelopes of segmented signal processed by short-time modified Hilbert transform | Squared mahalanobis distance classification | 99.43, 98.93, 99.13, 99.85, 98.62, 99.67 and 99.91 in the detection of MR, MS, ASD, NM, AS, AR and VSD, respectively |
| 2022 | See et al. [38] | PhysioNet 2016 | 2 | Shannon entropy and spectral entropy from three frequency bands | SVM | 82.50 |
| 2022 | Zhou et al. [39] | PhysioNet 2016 | 2 | — | Dense feature selection convolution network | 86.70 |
| 2021 | Shuvo et al. [40] | Yaseen dataset | 5 | Convolutional layers and max-pooling layers to extract time, frequency, and pattern features | CardioXNet (A CRNN network, including representation learning and sequence residual learning) | 88.09 |
| 2021 | Gelpud et al. [41] | PhysioNet 2016 | 2 | CWT scalogram of segmented sounds based on teager energy operator and autocorrelation | ResNet152 and VGG16 | ResNet152: 91.19
VGG16: 90.75 |
| 2022 | Arshad et al. [42] | PhysioNet 2016 | 2 | Mean value of each window of short-time power spectral density | Decision tree | 84.94 |
| 2021 | Duggento et al. [43] | PhysioNet 2016 | 2 | MFCC | Ad hoc multibranch, multiinstance artificial neural network | 97.00 |
| 2022 | Tian et al. [44] | PhysioNet 2016 | 2 | Extract gramian angular fields image features by ResNet2 and extract sequence features by MobileNet-LSTM | Combination of ResNet2 and MobileNet-LSTM | 97.99 |
|
