Removal of imaging artifacts in EEG during simultaneous EEG/fMRI recording: reconstruction of a high-precision artifact template

Functional magnetic resonance imaging (fMRI) induces coarse electromagnetic artifacts into the simultaneously recorded electroencephalogram (EEG). The problem in the signal processing framework is to model the underlying artifact, which is time-continuous, as a discretely sampled waveform. To build up an artifact template, the EEG sampling in relation to the phase of the imaging artifacts should be known. If the MR scanner and EEG sampling are not synchronized, this relation is not constant and a time adjustment of the template with the individual slice artifacts becomes essential. However, lack of synchrony opens up the possibility for approximating a high-precision and continuous artifact template by using the samples acquired from slightly different phases of the induced artifact. In this work, methodology for reconstructing such a template was developed using EEG data recorded simultaneously with fMRI at 3 T. A time-continuous cubic spline approximation was used as the slice artifact model. To overcome the problem of non-synchronized clocks, two methods were proposed to find the starting times of the slice artifacts at sub-sample precision. This approach yielded efficient imaging artifact reduction: the amplitude at the dominant frequency was attenuated by 55-70 dB (the median values over EEG channels) and the residual signal, at its best, was practically free from sharp transients even with 5000 Hz sampling frequency and without further residual artifact reduction algorithms. The presented methods may reduce the need for post-processing of the residual signal after the template subtraction and may help to preserve the EEG bandwidth.