Enhanced Motion Artifact Mitigation in ECG Signals using Nonlinear Autoregressive Networks with Cat Swarm Optimization and Fractional Calculus

The Motion artifacts in ECG data can lead to inaccurate circulatory state analysis. Recently, there has been growing interest in methods to mitigate motion distortion in ECG signals. One major challenge is to reduce motion artifacts without affecting the underlying ECG signal, as the motion artifacts and the ECG signal often overlap. Adaptive noise cancellers have proven effective in reducing motion artifacts, provided that an appropriate noise reference, which correlates with the noise in the ECG signal, is available. However, the correlation between motion distortion and the noise reference is not always consistent, and using an inappropriate noise reference can contaminate the ECG data. In this research, we present NARX_CSOFC, an advanced approach that significantly enhances the capabilities of the nonlinear autoregressive network with exogenous inputs (NARX) architecture by incorporating nonlinear combinations of input variables to globally estimate any nonlinear function. The method employs cat swarm optimization with fractional calculus (CSOFC) to find optimal solutions. The proposed NARX_CSOFC demonstrates substantial improvements in artifact reduction: achieving a 12 dB improvement for ECG artifacts, a 16 dB improvement for EMG artifacts, and a 15 dB improvement for random noise artifacts, as measured by Signal-to-Noise Ratio (SNR) and Mean Squared Error (MSE) compared to existing techniques.