Computational NeuroscienceA new informed tensor factorization approach to EEG-fMRI fusion

- Using PARAFAC2 as a multimodal BSS tool for fMRI analysis.
- Extracting information from EEG to detect post-movement beta rebound.
- Incorporating extracted information from EEG into PARAFAC2 to find active voxels.
- fMRI analysis and EEG-fMRI fusion using GLM and the proposed method.

BackgroundIn this paper exploitation of correlation between post-movement beta rebound in EEG and blood oxygenation level dependent (BOLD) in fMRI is addressed. Brain studies do not reveal any clear relationship between synchronous neuronal activity and BOLD signal. Simultaneous recording of EEG and fMRI provides a great opportunity to recognize different areas of the brain involved in EEG events.New methodIn order to incorporate information derived from EEG signals into fMRI analysis a specific constraint is introduced in this paper. Here, PARAFAC as a variant of tensor factorization, exploits the data changes in more than two modes in order to reveal the information about the fMRI BOLD and its time course simultaneously. In addition, various constraints can be applied during the alternating process for estimation of its parameters.ResultsThe achieved results from extensive set of experiments confirm effectiveness of the proposed method to detect the brain regions responsible for beta rebound. Moreover, fMRI-only and EEG-fMRI analysis using PARAFAC2 illustrate correct expected activities in the brain area.Comparison with existing methodsThe advantages of the proposed method are revealed when comparing the results with those of obtained using general linear model (GLM) which is a well-known model-based approach.ConclusionsThe proposed method is a semi-blind decomposition technique which employs PARAFAC2 without relying on a predefined time course. The achieved results indicate that this approach can pave the path for multi-task analysis in BCI applications.