Table 3.
Summary of the previous heart sound classification works. Se: sensitivity, Sp: specificity and Acc: accuracy.
| Author | Database | Recording length | Classification method | Features | Se (%) | Sp (%) | Acc (%) | Notes on database |
|---|---|---|---|---|---|---|---|---|
| (Akay et al., 1994) | 42 normal and 72 CAD patients | Each 10 cycles | ANN | Wavelet | 85 | 89 | 86 | 30 training, 82 test |
| (Liang and Hartimo, 1998) | 40 normal and 45 pathological children | Each 7–12 s | ANN | Wavelet | 80 | 90 | 85 | 65 training, 20 test |
| (Uguz, 2012a) | 40 normal, 40 pulmonary and 40 mitral stenosis | -- | ANN | Wavelet | 100 | 95.24 | 98.33 | 50–50 train-test split |
| (Bhatikar et al., 2005) | 88 innocent murmurs and 153 pathological murmurs | Each 10–15 s | ANN | Frequency | 83 | 90 | -- | 188 training, 53 test |
| (Sepehri et al., 2008) | 36 normal and 54 pathological | Each 10 s | ANN | Frequency | 95 | 93.33 | -- | 40 training, 50 test |
| (Ahlstrom et al., 2006) | 7 normal, 23 aortic stenosis and 6 mitral regurgitation | Each 12 cycles | ANN | Complexity | -- | -- | 86 | Cross-validation |
| (De Vos and Blanckenberg, 2007) | 113 normal and 50 pathological | Each 6 cycles | ANN | Time-frequency | 90 | 96.5 | -- | Cross-validation |
| (Uguz, 2012b) | 40 normal, 40 pulmonary and 40 mitral stenosis | -- | ANN | Time-frequency | 90.48 | 97.44 | 95 | 50–50 train-test split |
| (Ari et al., 2010) | 64 patients (normal and pathological) | Each 8 cycles | SVM | Wavelet | -- | -- | 86.72 | 50–50 train-test split |
| (Zheng et al., 2015) | 40 normal and 67 pathological | -- | SVM | Wavelet | 93.48 | 98.55 | 97.17 | Cross-validation |
| (Patidar et al., 2015) | Total 4,628 heart cycles, 626 normal and 4,002 pathological | -- | SVM | Wavelet | 98.8 | 99.3 | 98.9 | 80% training, 20% test |
| (Maglogiannis et al., 2009) | 38 normal and 160 heart valve disease patients | -- | SVM | Frequency | 87.5 | 94.74 | 91.43 | Cross-validation |
| (Gharehbaghi et al., 2015) | 30 normal, 26 innocent and 30 aortic stenosis | Each 10 s | SVM | Frequency | 86.4 | 89.3 | -- | 50–50 train-test split |
| (Wang et al., 2007) | 20 normal and 21 murmurs patients | -- | HMM | Signal amplitude, STFT and MFCC | ≥95.2 | ≥95.3 | -- | No separate training and test |
| (Chauhan et al., 2008) | 20 normal and 21 murmurs patients | -- | HMM | Signal amplitude, STFT and MFCC | -- | -- | 99.21 | No separate training and test |
| (Saracoglu, 2012) | 40 normal, 40 pulmonary and 40 mitral stenosis | HMM | DFT and PCA | 95 | 98.8 | 97.5 | 50–50 train-test split | |
| (Bentley et al., 1998) | Unspecified size: native and prosthetic heart valves patients | -- | kNN | Wavelet | -- | -- | 100 for native and 87 for prosthetic | Cross-validation |
| (Quiceno-Manrique et al., 2010) | 16 normal and 6 pathological | -- | kNN | Time-frequency | -- | -- | 98 | Cross-validation |
| (Avendano-Valencia et al., 2010) | 26 normal and 19 pathological | -- | kNN | Time-frequency | 99.56 | 98.45 | 99.0 | Cross-validation |