Noise reduction in combined EEG/fMRI using a vector beamformer

Introduction: BOLD fMRI achieves excellent spatial localization of brain activity, however the latency and longevity of the haemodynamic response means that fMRI suffers from poor temporal resolution. Conversely, electroencephalography (EEG) directly measures the electrical potentials generated by neuronal activity, and therefore offers excellent temporal resolution, but the ill-posed EEG inverse problem and inhomogeneous conductivity profile in the head mean that its spatial resolution is limited. For these reasons, simultaneous application of fMRI and EEG represents an attractive way of imaging brain function with high spatio-temporal resolution. Simultaneous application is difficult due to the EEG artifacts caused by the MR scanner. Techniques are available to remove such artifacts [1,2], however, even after correction, noise remains in the data, particularly at high field. Here, we demonstrate a vector beamformer [2] approach that, if applied in addition to standard correction techniques, further reduces artifacts in EEG/fMRI. We show that, as well as significantly reducing noise in EEG data, this approach allows electrical source localization and extraction of virtual electrode timecourses. Theory: Assume that m(t) represents an n dimensional column vector of measurements made at n EEG electrodes at time t in response to a current dipole of strength Q at location r. An estimate of Q(r, t) can be made by a weighted sum of sensor measurements such that ( ) ( ) ( ) t t m r W r Q = , ˆ