ICA去除EEG中眼动伪差和工频干扰方法研究

Study on Applying Independent Component Analysis to Remove Blink Artifacts and Power Noise in EEG

眼动伪差和工频干扰是临床脑电图 (EEG)中常见噪声 ,严重影响其有用信息提取 .本文尝试采用独立分量分析 (IndependentComponentAnalysis,ICA)方法分离EEG中此类噪声 .通过对早老性痴呆症 (Alzheimerdisease,AD)患者临床EEG信号 (含眼动伪差和混入工频干扰 ,信噪比仅 0dB)作ICA分析 ,比较了最大熵 (Infomax)和扩展最大熵(ExtendedInfomax)ICA算法的分离效果 ,证实虽然最大熵算法可以分离出眼动慢波 ,但难以消除工频干扰 ,为此需采用扩展的最大熵算法 ;并知ICA方法在极低信噪比时也有较好的抗干扰性 ,且在处理非平稳信号时有好的鲁棒性 ;文中还结合近似熵 (approximateentropy ,ApEn)分析说明利用ICA去除干扰后有助于恢复和保持原始EEG信号的非线性特征 .研究结果表明ICA方法在生物医学信号处理中具有潜在的重要应用价值 ,值得深入研究和推广 .

Blink artifacts and power noise are constantly found: to strongly influence the acquisition and analysis of EEG signals. In this paper, by comparing the efficiencies of two ICA algorithms-Infomax-ICA and Extended-Infomax-ICA methods in extracting blink artifacts and power noise in the EEG signals, it was shown that ICA algorithms were insensitive to disturbance in the conditions of low signal-noise-ratio, and ICA algorithms demonstrated a strong robustness in processing non-stationary signals. Though blink slow waves could be extracted by infomax algorithm, but power noise was unlikely to be removed by it. Therefore, Extended-Infomax ICA algorithm should be used. By applying Extended-Infomax algorithms, blink artifacts and power noise contained in the 16-channel EEG signals of Alzheimer-disease patients were removed successfully (the lowest signal-noise-ratio for power noise can be -40 dB). Meanwhile, it proved by calculating approximation entropy (ApEn) that ICA algorithms could preserve the nonlinear characteristics of EEG after removing the interference.