Relevance and Objectives. Contact noise remains a key issue for the quality of ECG signals recorded by minicardiographs with finger electrodes due to high contact resistance and its variability. Traditional methods, including linear filters and EEMD, are either insufficiently effective for non-stationary noise or are too resource-intensive for mobile devices. Hybrid approaches based on EEMD with wavelet thresholding (WT) offer a balance between efficiency and computational load, allowing for better preservation of signal morphology, including the low-amplitude P-wave. The objective of this study was to develop a method for optimizing the parameters of the hybrid EEMD+WT algorithm to maximize contact interference suppression in ECG signals from a minicardiograph with finger electrodes on affordable smartphones, while preserving diagnostically significant characteristics and improving P-wave visibility compared to existing hybrid methods (e.g., SG+EEMD).

Materials and Methods. The study was based on 10-second ECG recordings obtained with the "Serdechko" minicardiograph with finger electrodes and transmitted via the smartphone’s audio channel. The sampling frequency was 250 Hz, and the bandwidth was 0.5−25 Hz. Twenty patients (10 men, 10 women) aged 45−85 years participated in the study. A total of 200 recordings with contact interference were obtained (maximum peak-to-peak 0.6 mV, average power 0.05 mW²). The EEMD+WT algorithm is implemented in Python using the PyEMD and SciPy libraries. EEMD first decomposes the signal into IMFs with white noise (signal amplitude 0.2 σ, 10 ensembles), then WT is applied to the high-frequency IMFs. Combinations of ensembles (5−20) and WT levels (3−6) were tested. Evaluation criteria: noise power reduction (target ≥90%), SSIM (target ≥0.95), and processing time (target ≤1000 ms).

Results. Optimal parameters: 10 EEMD + WT ensembles (level 4). This achieves a noise power reduction of 94%, SSIM = 0.95, and a processing time of 750 ms. The EEMD+WT hybrid outperforms SG+EEMD in noise suppression (by 4%) and a 27% reduction in processing time. Preservation of diagnostic data: R-wave amplitude measurement error is 1.2%, P-wave amplitude measurement error is 4.1%. The method copes better with non-stationary interference, improving P-wave visibility in 95% of cases.

Conclusions. The hybrid EEMD algorithm with wavelet thresholding and optimized parameters provides effective suppression of contact interference while fully preserving ECG morphology and diagnostic parameters, with computational efficiency suitable for low-cost mobile devices. This improvement over SG+EEMD makes it preferable for integration into portable cardiographs.