Wavelet-based ECG-derived Respiration Denoising

We propose a wavelet-based EDR (Electrocardiogram derived respiration) denoising algorithm. When a QRS complex of ECG is misdetected, EDR is abruptly corrupted by huge noise. To mitigate such noise, we employed wavelet transform and likelihood functions (Gaussian mixture model and Laplace distribution). Likelihood based hard thresholding was performed for wavelet coefficients and it effectively eliminated noise in EDR signal. To verify the algorithms, we used the MIT-MIMIC open source data with simulated spike random noise. Most correlation coefficients and mean absolute errors of filtered EDRs were significantly higher and lower than those of contaminated EDRs ($p < 0.0001$), respectively. Since EDR can be used to estimate not only respiratory rate but also tidal volume, we expect that the proposed method can enhance the reliability and utility of IoMT devices with ECG.