A unified canonical correlation analysis-based framework for removing gradient artifact in concurrent EEG/fMRI...

Artifacts cause distortion and fuzziness in electroencephalographic (EEG) signal and hamper EEG analysis, so it is necessary to remove them prior to the analysis. Particularly, artifact removal becomes a critical issue in experimental protocols with significant inherent recording noise, such as mobile EEG recordings and concurrent EEG-fMRI acquisitions. In this paper, we proposed a unified framework based on canonical correlation analysis for artifact removal. Raw signals were reorganized to construct a pair of matrices, based on which sources were sought through maximizing autocorrelation. Those sources related to artifacts were then removed by setting them as zeros, and the remaining sources were used to reconstruct artifact-free EEG. Both simulated and real recorded data were utilized to assess the proposed framework. Qualitative and quantitative results showed that the proposed framework was effective to remove artifacts from EEG signal. Specifically, the proposed method outperformed independent component analysis method for mitigating motion-related artifacts and had advantages for removing gradient artifact compared to the classical method (average artifacts subtraction) and the state-of-the-art method (optimal basis set) in terms of the combination of performance and computational complexity.