Motion and Ballistocardiogram Artifact Removal for Interleaved Recording of EEG and EPs during MRI Giorgio Bonmassar, 1 Patrick L. Purdon, Iiro P. Ja ¨a ¨ skela ¨ inen, Keith Chiappa, Victor Solo, Emery N. Brown, and John W. Belliveau NMR Center, Massachusetts General Hospital, Harvard Medical School, and A. Martinos Center for Biomedical Imaging, Charlestown, Massachusetts 02129 Received February 2, 2001 Artifacts generated by motion (e.g., ballistocardiac) of the head inside a high magnetic field corrupt record- ings of EEG and EPs. This paper introduces a method for motion artifact cancellation. This method is based on adaptive filtering and takes advantage of piezoelec- tric motion sensor information to estimate the motion artifact noise. This filter estimates the mapping be- tween motion sensor and EEG space, subtracting the motion-related noise from the raw EEG signal. Due to possible subject motion and changes in electrode im- pedance, a time-varying mapping of the motion versus EEG is required. We show that this filter is capable of removing both ballistocardiogram and gross motion artifacts, restoring EEG alpha waves (8 –13 Hz), and visual evoked potentials (VEPs). This adaptive filter outperforms the simple band-pass filter for alpha de- tection because it is also capable of reducing noise within the frequency band of interest. In addition, this filter also removes the transient responses normally visible in the EEG window after echo planar image acquisition, observed during interleaved EEG/fMRI recordings. Our adaptive filter approach can be imple- mented in real-time to allow for continuous monitor- ing of EEG and fMRI during clinical and cognitive studies. © 2002 Elsevier Science (USA) Key Words: adaptive filtering; ballistocardiogram (BCG); interleaved VEPs and fMRI; visual evoked po- tentials; VEP; EEG; MRI; brain mapping; epilepsy. INTRODUCTION Simultaneous recording of EEG and fMRI is an im- portant, emerging tool in functional neuroimaging that combines the high spatial resolution of fMRI with the high temporal resolution of EEG (Belliveau et al., 1993; Dale and Sereno 1993; Ives et al., 1993; Belliveau et al., 1995; George et al., 1995; Menon et al., 1997; Bonmas- sar et al., 1999; Hill et al., 1999; Schomer et al., 2000; Bonmassar et al., 2001a, Bonmassar et al., 2001b). Applications of this technique include recordings of visual and auditory evoked potentials, studies of sleep, and EEG-triggered fMRI to study epilepsy and other neurological disorders. A fundamental limitation of this combined technique is the noise introduced in the EEG due to motion within the magnetic field, either from cardiac pulsation (ballistocardiogram) or from head movements (Huang-Hellinger et al., 1995). The ballistocardiogram noise may obscure EEG activity at alpha frequencies (8 –13 Hz) and below, with ampli- tudes often in excess of 150 V at 1.5 T field strength, much larger than the alpha waves seen in most sub- jects (50 V). In some subjects the ballistocardiogram noise is not visible. Head rotations and translations, present in longer recordings or in recordings of pa- tients with certain neurological disorders, result in even larger disturbances to the EEG. In evoked-poten- tial studies, these large-amplitude disturbances result in rejection of epochs, increasing the duration of such experiments due to lost data, and making recordings from subjects who possess large ballistocardiogram waveforms. In EEG-triggered fMRI studies of epilepsy, it is essential to have clinically interpretable EEG from which to identify epileptic events in real-time (Schomer et al., 2000). Ballistocardiogram and motion-induced noise can make these epileptic events difficult to detect with precision (Krakow et al., 1999) and can mimic epileptic discharge (Hill et al., 1995). In sleep studies, identification of delta (0.5– 4 Hz) (Kandel et al., 2000) and alpha waves is essential, yet the ballistocardio- gram artifact directly obscures these frequency ranges. Because these sources of noise are the direct result of electromagnetic induction within the static magnetic field, the problem becomes worse at higher fields. One method for removing the ballistocardiogram noise is to subtract an average ballistocardiogram 1 To whom correspondence and reprint requests should be ad- dressed. Fax: (617) 726 7422. E-mail: giorgio@nmr.mgh.harvard.edu. NeuroImage 16, 1127–1141 (2002) doi:10.1006/nimg.2002.1125 1127 1053-8119/02 $35.00 © 2002 Elsevier Science (USA) All rights reserved.