A Highly Robust EMG Noise Filtering Method Based on Synchrosqueezing Generalized Phase-Shifting S-Transform and Image Connectivity Threshold

Abstract:
Accurate electrocardiogram (ECG) signal acquisition is essential for cardiac health assessment and the diagnosis of cardiovascular conditions. While portable monitoring devices enable real-time ECG signal collection in non-clinical environments, the recorded signals are often heavily contaminated by electromyographic (EMG) noise. Due to the broad spectral overlap between EMG noise and the effective ECG signal, traditional frequency-domain filtering methods face significant challenges in achieving clean signal separation. To address this challenge, this paper proposes a novel twostage EMG noise suppression approach for ECG signals. The method leverages the synchrosqueezing generalized phaseshifting S-transform (SS-GPST) for high-resolution time-frequency representation, followed by image processing techniques based on pixel connectivity thresholding. A systematic comparative analysis was conducted to evaluate the performance differences among wavelet transform (WT), ensemble empirical mode decomposition (EEMD), and SS-GPST in EMG noise suppression. Experimental validation using standard MIT-BIH arrhythmia database data demonstrates that the proposed algorithm exhibits superior denoising performance under varying EMG noise contamination intensities. Quantitative results show that, even under severe noise conditions (input SNR < –5dB), the proposed method significantly outperforms the baseline techniques, achieving an average SNR improvement of 16.63dB, reducing RMSE to 0.087, and attaining 90.21% waveform similarity with the clean reference signal. These results demonstrate the effectiveness and robustness of the method in preserving ECG morphology and improving signal fidelity in challenging acquisition environments.