A Review of EEG Artifact Removal Techniques for Brain-Computer Interface

Electroencephalography (EEG) is a vital technique for analysing brain function; however, signals are often contaminated by artifacts from eye movements, muscle activity, and cardiac rhythms. These distortions interfere with the true neural signal, complicating accurate interpretation. Effective artifact-removal techniques are therefore essential to enhance the reliability of EEG analysis in both research and clinical applications. This review systematically examines EEG artifact removal methods published in open-access, peer-reviewed journals between 2010 and 2025. The selected studies were chosen based on methodological clarity, quantitative evaluation using metrics such as Peak Signal-to-Noise Ratio (PSNR) and Correlation Coefficient (CC), and validation on benchmark datasets including PhysioNet, EEGdenoiseNet, and BCI Competition IV. The analysis compares traditional signal-processing approaches (manual inspection, notch and bandpass filtering, spatial and regression techniques) with advanced decomposition methods such as Independent Component Analysis (ICA), Discrete Wavelet Transform (DWT), and Empirical Mode Decomposition (EMD/CEEMDAN). Traditional filters efficiently remove specific frequency noise but may distort neural components. Advanced decomposition-based methods improve artifact suppression and signal fidelity but are computationally intensive and parameter-sensitive. Hybrid frameworks, integrating signal-processing and deep-learning methods such as VMD–SWT–CCA, CNN–LSTM–EMG, and CNN–LMS, achieve superior performance, improving PSNR by 3–6 dB and CC by 0.04–0.07 compared to single methods. No single approach can address all artifact types; however, hybrid and data-driven models show the highest potential for real-time, accurate, and robust EEG artifact removal, paving the way for more reliable and clinically relevant EEG analysis.