Optimized adaptive filtering using genetic algorithm and signal quality index for practical motion artefact reduction in ambulatory ECG

Motion artefact (MA) is the most challenging noise to remove during electrocardiogram (ECG) signal pre-processing. Its presence can cause incorrect detection of QRS complexes, P and T wave distortion, and, as a result, misdiagnosis. While adaptive noise cancelers are a widely used practical solution for MA removal, their performance depends heavily on selecting optimal parameters, such as step size and filter length. In this study, we propose an automated approach for optimizing these parameters to enhance MA removal. We introduce a two-step knowledge-based genetic algorithm (k-GA) framework comprising (1) knowledge acquisition and (2) filter optimization. Instead of using mean squared error as the fitness function, the k-GA employs template signal quality indices (tSQI), making it suitable for real-world ECG signals. We explored the k-ga with the sign LMS (k-gaSLMS), sign-sign LMS (k-gaSSLMS), and sign data LMS (k-gaSDLMS). We simulated noisy ECG signals using the MIT-BIH arrhythmia database and the Noise Stress Test Database. The BRNO University of Technology ECG quality database provides real-world noisy ECGs and accelerometer data that are pre-processed to serve as the input reference of the adaptive filter. Compared to the LMS, SDLMS, and NLMS, our results show that the proposed method (k-gaSDLMS) gives higher improvement in signal-to-noise ratio (14.36~15.82 dB), lower root-mean-square-error (0.04~0.06), and higher correlation coefficient (97~99%) on simulated data. For real-life data, the tSQI passes from 0.49 to 0.63, showing that the k-gaSDLMS can automatically select filtering parameters and efficiently remove MA. The average execution time (~0.30 s per 6−s segment) supports its integration into near-real-time or offline wearable ECG applications. Thus, the proposed method enables more reliable signal pre-processing in ambulatory monitoring and may contribute to improved downstream diagnostic accuracy.