The evaluation of seismocardiogram signal pre-processing using hybridized variational mode decomposition method

This study aims to determine the performance of variational mode decomposition (VMD) combined with detrended fluctuation analysis (DFA) as a hybrid framework for extracting seismocardiogram and respiration signals from simulated single-channel accelerometry data and removing its contained noise. The method consists of two consecutive layers of VMD that each contribute to extracting respiration and SCG signal respectively. DFA is utilized to determine the number of modes produced by VMD and select the most appropriate modes to be the constituents of the reconstructed signal based on the Hurst exponent value thresholding. This hybridized VMD successfully extracted respiration and SCG signal with minimal mean absolute error value (0.516 and 0.849, respectively) and boosted the SNR to 2 dB and 4 dB, respectively in heavily noise-interfered conditions. This method also outperformed other empirical mode decomposition strategies and exhibits short computational time. Two main drawbacks exist in this framework, i.e. the determination of balancing parameter ( $$ ) that is still conducted manually and the magnitude shifting phenomenon. In conclusion, the hybridized VMD shows an outstanding performance in denoising and extracting respiration and SCG signals from a single input that combines them and generally is impured by noise.