Blind separation of ECG signals from noisy signals affected by electrosurgical artifacts

Electrocardiogram (ECG) signal monitoring is crucial in the operating room. During surgery, when using an electrosurgical unit (ESU), the ECG signal is corrupted by strong interference generated by the ESU, so-called Electrosurgical Artifacts (EAs). The objective of this study is to enhance the infected ECG signal and to remove the impact of the EAs. Motivated by the fact that the artifacts’ energy is greater than that of the ECG signal, we propose a new algorithm to select the intrinsic mode function (IMFs) based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) in combination with a normalized least mean square (NLMS) adaptive filter. More specifically, CEEMDAN is firstly applied to the observed ECG signals. Then, several IMFs are reconstructed and finally, the EAs and the de-noised ECG signal are estimated. To further improve the estimated ECG signal, a NLMS adaptive filter is applied by exploiting the noise estimated by the CEEMDAN-based algorithm. Our experimental results are obtained using real ECG signals. To illustrate the effectiveness of the proposed algorithm, we compare our results with those obtained by applying the thin singular value decomposition, an adaptive SVD or wavelets, all these methods are mixed with an adapter filter. The accuracy of all method is assessed using different performance metrics, such as MSE and improvement SNR. The proposed method outperforms all other methods. Indeed, it improves the SNR by an average of 7 dB and provides an ECG signal with a regular rhythm.