Predicting EEG single trial responses with simultaneous fMRI and Relevance Vector Machine regression

F. de Martino*, A.W. de Borst, G. Valente, R. Goebel, E. Formisano

*Corresponding author for this work

Research output: Contribution to journalArticleAcademicpeer-review

118 Downloads (Pure)


The combination of electroencephalography (EEG) and functional Magnetic Resonance Imaging (fMRI) has been proposed as a tool to study brain dynamics with both high temporal and high spatial resolution. Integration through EEG-fMRI trial-by-trial coupling has been proposed as a method to combine the different data sets and achieve temporal expansion of the fMRI data (Eichele et al., 2005). To fully benefit of this type of analysis simultaneous EEG-fMRI acquisitions are necessary (Debener et al., 2006). Here we address the issue of predicting the signal in one modality using information from the other modality. We use multivariate Relevance Vector Machine (RVM) regression to "learn" the relation between fMRI activation patterns and simultaneously acquired EEG responses in the context of a complex cognitive task entailing an auditory cue, visual mental imagery and a control visual target. We show that multivariate regression is a valuable approach for predicting evoked and induced oscillatory EEG responses from fMRI time series. Prediction of EEG from fMRI is largely influenced by the overall filtering effects of the hemodynamic response function. However, a detailed analysis of the auditory evoked responses shows that there is a small but significant contribution of single trial modulations that can be exploited for linking spatially-distributed patterns of fMRI activation to specific components of the simultaneously-recorded EEG signal. (PsycINFO Database Record (c) 2011 APA, all rights reserved) (journal abstract)
Original languageEnglish
Pages (from-to)826-836
Issue number2
Publication statusPublished - 1 Jan 2011


Dive into the research topics of 'Predicting EEG single trial responses with simultaneous fMRI and Relevance Vector Machine regression'. Together they form a unique fingerprint.

Cite this