Review Article
A Review of Computer-Aided Heart Sound Detection Techniques
Table 2
Segmentation methods of PCG signals.
| Year | Author | Segmentation method | Dataset | Result |
| 2019 | Giordano and Knaflitz [16] | Envelope-based technique | Sample population of 24 healthy subjects over 10-min-long simultaneous phonocardiography recordings | F1 of 99.2% | 2019 | Oliveira et al. [17] | HSMM-GMM | PhysioNet [18], PASCAL [19] and a pediatric dataset composed of 29 heart sounds | F-score of 92% | 2019 | Kamson et al. [20] | HSMM | Training-set-a of 2016 PhysioNet/computing in cardiology challenge | Sensitivity | + | F1 | 98.28 | 98.45 | 98.36 | 2019 | Renna et al. [21] | HSMM-CNN | PhysioNet | Sensitivity: 93.9% | 2018 | Liu et al. [22] | Time-domain analysis, frequency-domain analysis and time-frequency-domain analysis | Heart sound & Murmur library of UMich | Sensitivity: 98.63% | 2018 | Belmecheri et al. [23] | Correlation coefficients matrix | A database of 21 clean heart sounds | Sensitivity: 76% | 2018 | Alexander et al. [24] | HMM | 3240 PCG recordings from PhysioNet and PASCAL | Sensitivity | Specificity | | 90.3% | 89.9% | | 2017 | Babu et al. [25] | VMD | Database: | Sensitivity | + | Accuracy | PhysioNet | 98.90 | 96.07 | 95.14 | PASCAL | 99 | 100 | 99 | Michigan [26] | 100 | 100 | 100 | eGeneralMedical [27] | 100 | 100 | 100 | Real-time PCG signals | 100 | 97.08 | 97.08 | 2017 | Varghees et al. [28] | EWT | PhysioNet, PASCAL, Michigan, eGeneralMedical and real-time PCG signals | Sensitivity | Pp | OA | 94.38% | 97.25% | 91.92% | 2017 | Liu et al. [29] | HSMM | More than 120 000 s of heart sounds recorded from eight independent heart sound databases | F1 of 98.5% | 2016 | Thomas et al. [30] | Fractal decomposition (FD) | Michigan (23 different heart sounds and 6 patients’ recordings done in a real clinical environment) | Sensitivity | + | DER | 96.97 | 99.58 | 3.55 | 2016 | Springer et al. [31] | HSMM | 405 synchronous 30–40 s PCG and ECG recordings from 123 deidentified adult patients | F1 of 95.63 ± 0.85% | 2015 | Salman et al. [32] | Peak intervals pattern | 1089 cycles from 62 set of normal and abnormal signals | Correct cycle detected rate of 83.38% |
|
|