Table 2.
Denoising methods for epileptic EEG signals.
| Author | Dataset | Denoise methods | Type | Result | Advantages | Disadvantages |
|---|---|---|---|---|---|---|
| Zhou et al. (2021) | Non-public | DWT, dynamic thresholding | DWT | precision:86.8% sensitivity: 82.7% | Window is Adjustable (Chen et al., 2024) | Basis functions are not adaptive (Sharma, 2017; Huang et al., 1998) |
| Yedurkar and Metkar (2020) | Non-public CHB-MIT | DWT, adaptive filtering | accuracy: 86.66% Precision:88.88% | |||
| Parija et al. (2020) | Bonn | EMD | EMD | Accuracy:100% | Basis functions are adaptive (Das et al., 2024) | modal aliasing (Song et al., 2023) High time complexity |
| Moctezuma and Molinas (2020) | CHB-MIT | Minkowski Distance, EMD | accuracy: 93% | |||
| Karabiber Cura et al. (2020) | Non-public | EEMD | 1.5% improvement in classification accuracy | Reducing modal aliasing | Noise residue (Lan et al., 2024) High time complexity | |
| Hassan et al. (2020) | Bonn | CEEMDAN, NIG parameters | Over 97% accuracy, sensitivity, specificity | Reducing noise residue | High time complexity | |
| Bari and Fattah (2020) | Bonn | CEEMDAN | Accuracy reduced by 1%, computational complexity and time complexity reduced | |||
| Liu et al. (2022) | Bonn Freiburg | Correlation,VMD | Sensitivity and specificity of more than 95% | Reduced modal aliasing and reduced computational complexity | Slow parameter selection and poor generalization | |
| Peng et al. (2021) | BERN-BARCELONA | EVMD | Accuracy, sensitivity and specificity increased by more than 3%. | |||
| De Vos et al. (2011) | Non-public | ICA | BSS | The number of false alarms has been reduced by almost four times | No need to know the signal artifact type (Uddin et al., 2023), Lower time complexity than EMD | Reference signals required (Xu et al., 2024) |
| Islam et al. (2020) | Non-public | Infomax ICA | Accuracy can be improved by 24% | |||
| Becker et al. (2015) | Non-public | deflation ICA | Reduce computational complexity by a factor of 10 | |||
| Sardouie et al. (2014) | Non-public | TF-GEVD, TF-DSS | Better results than CCA and ICA | |||
| Qiu et al. (2018) | Bonn | DSAE | Deep learning | Performance improved by 8.19% | Less residual noise | High computational complexity, High time complexity |
| Lopes et al. (2021) | EPILEPSIAE | DCNN | Evaluation indicators are better than Infomax ICA-MARA | |||
| Jana et al. (2022) | CHB-MIT Bonn | DWT-EMD | Fusion methods | Accuracy can be improved by 19.58% | Less residual noise | High time complexity |
| Du et al. (2024) | New Delhi | CEEMDAN-CWT | The signal-to-noise ratio (SNR) can be increased by 1.0567 dB and the root mean square error (RMSE) can be reduced by 0.1045 |