Correlated coupled matrix tensor factorization method for simultaneous EEG-fMRI data fusion

Abstract

ObjectiveFusion of EEG and fMRI data provides complementary information about the brain functions. Thus, data fusion should employ all dimensions of data to both extract the shared information, and distinguish the dissimilarities of modalities. MethodIn this paper, we present a new method based on Advanced Coupled Matrix Tensor Factorization (ACMTF) method to decompose the EEG tensor and fMRI matrix. We remove the confining assumption of identical shared components in ACMTF and propose an approach to maximize the correlation between the shared components of EEG and fMRI in the common mode, while removing the smearing effect of hemodynamic response function on fMRI data. This method is called Correlated Coupled Matrix Tensor Factorization (CCMTF). ResultsTwo simulation scenarios with different noise levels are performed to evaluate the performance of the proposed CCMTF method. The simulation results show 30.1% and 23.38% average improvement of Match Scores over ACMTF and N-way Partial Least Square (N-PLS) methods, respectively. The proposed method is also applied to a real dataset of an auditory oddball paradigm. The results of real data analysis show that the proposed CCMTF method estimates the unknown number of shared components between EEG and fMRI data, with p-value less than 0.007. Conclusion and SignificanceIn addition to the improvement of match score, the simulation results demonstrate that CCMTF has higher performance than ACMTF for fusion of EEG and fMRI datasets in case of extracting the dissimilarities in modalities, and it provides more convenient interpretation of the results, in comparison to N-PLS method.