Development of a technological platform for simultaneous EEG-fMRI data integration

Multimodal imaging based on simultaneous electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) is a viable approach for investigating human brain function. EEG provides a direct measure of oscillatory electrophysiological activity, which originates from the firing of cortical neurons. It has high temporal resolution, on the scale of milliseconds. fMRI has proved to be very accurate in displaying brain activations with a spatial resolution of a few millimeters. It detects the slow hemodynamic fluctuations, which are measured using the Blood Oxygenation Level Dependent (BOLD) contrast, and are only indirectly associated with neuronal activity. Accordingly, the integration of EEG and fMRI allows the visualization of the same phenomenon, the brain activation, through two different spatial-temporal windows of analysis, each of them able to show particular features. Though, the EEG-fMRI integration is methodologically challenging, as it may compromise data quality and raise concerns on subject safety. Growing efforts have been made to tackle these challenges, resulting in the introduction of specifically magnetic resonance (MR)-compatible designed EEG instrumentation, and the development of more and more advanced techniques for dealing with artifactual sources. Despite consistent improvements over the years, the EEG signals recorded during simultaneous fMRI scanning are still contaminated by strong artifacts, induced by the interactions between the MR environment, the EEG equipment and the participant. For an accurate characterization of the measured brain signals, they should be removed before any further analysis. The gradient artifact, generated by the fMRI acquisition process, has a massive effect, but can be relatively easily removed due to its high reproducibility. Gradient artifact aside, the EEG recordings are still heavily corrupted by the presence of the ballistocardiographic (BCG) artifact, which is associated with the ongoing cardiac activity of the participant. Despite some methods to remove the BCG exist, this processing step is very challenging due to the complex BCG spatio-temporal dynamics. Once EEG data are cleaned from the MR-related artifacts, they are given as an input for conventional, out of MR scanner, EEG analyses. This includes further pre-processing steps and, eventually, EEG source localization. For the latter, the major issue raises from the definition of the volume conduction model, constructed by the integration of information about the subject’s head anatomy, tissues biophysical, e.g. electrical and physical, properties, and position of the EEG sensors over the scalp. This model is meant to theoretically establish how brain sources can generate specific distributions of EEG potentials, and its accuracy is strongly influenced by the correctness of the information given in input. Thus, the availability of clean EEG data and the definition of a reliable volume conduction model are essential requirements for accurate EEG neural sources imaging. Once EEG data are cleaned from the MR-related artifacts, they are given as an input for conventional, out of MR scanner, EEG analyses. This includes further pre-processing steps and, eventually, EEG source localization. For the latter, the major issue raises from the definition of the volume conduction model, constructed by the integration of information about the subject’s head anatomy, tissues biophysical, e.g. electrical and physical, properties, and position of the EEG sensors over the scalp. This model is meant to theoretically establish how brain sources can generate specific distributions of EEG potentials, and its accuracy is strongly influenced by the correctness of the information given in input. Thus, the availability of clean EEG data and the definition of a reliable volume conduction model are essential requirements for accurate EEG neural sources imaging.