Event Abstract

Back to Event

EEG-fMRI fusion on the cortical surface using Coupled Tensor-Matrix Factorization: A simulation study

Esin Karahan-Senvardar1*, Adil Deniz Duru2, Pedro A. Valdes-Sosa3 and Ahmet Ademoğlu1, 4
  • 1 Bogazici University, Institute of Biomedical Engineering, Türkiye
  • 2 Istanbul University, Hulusi Behçet Life Sciences Research Center, Türkiye
  • 3 Cuban Neuroscience Center, Cuba
  • 4 Istanbul Sehir University, Türkiye

Although it is a challenging problem in neuroimaging to merge the data coming from different modalities on a common spatial domain, the integration of EEG and fMRI offers us an opportunity to reveal the complex dynamics of brain functions and neuronal interactions (Bayram et al., 2011).

In this study, a novel EEG-fMRI fusion approach based on multilinear methods is developed and applied to simulated data. We treat EEG data as a three-way array with temporal, spectral and spatial dimensions. We perform coupled tensor-matrix factorization (CTMF) to obtain common spatial signatures between fMRI and spectral EEG data (Acar et al., 2011). The EEG inverse problem is also incorporated into the merging process to obtain a common image on the cortical surface. We improve the spatial signature estimation by using the alternating least squares algorithm in a hierarchical manner where the noise and factor covariance are also exploited. Unlike conventional CTMF algorithms where a single dimension is considered to be fully coupled between two datasets, we project part of the datasets on a common and part on discriminative subspaces (Liu et al., 2013). This enables us to deal with the cases in which EEG and fMRI sources differ. By this way, our proposed algorithm is able to show both coupled and uncoupled responses at the same time.

In the simulation, we generated EEG/fMRI data from one deep and two superficial sources. Deep source is common to both modalities and each of the superficial sources is specific to one modality. Results of the simulation show that CTMF algorithm with inverse problem successfully localizes the sources. (Maximum values of the fMRI spatial components overlap with the real sources; common spatial EEG component overlaps with the real source and the distance between the maximum value of the uncommon EEG spatial component with the real one is 11 mm.) Also, algorithm correctly identifies common and distinct sources (See the figure).

Brain activity reflected in different domains is integrated on the same spatial scale by this approach. Our current research focuses on applying the method on real data.


Figure 1: Three different source locations are chosen on the cortical surface. Deep source is located in middle temporal gyrus and superficial source specific to fMRI is on the right superior-frontal cortex and superficial source specific to EEG is on the left superior-frontal cortex. fMRI temporal pattern is constituted by convolving hemodynamic function with a boxcar. EEG source signal is generated from sinusoids oscillating at 4 and 12 Hz. Channel EEG is obtained by projecting source signal onto the sensor space with lead field matrix. First two rows shows the fMRI spatial and temporal components found from CTMF algorithm. Last two rows are the EEG components. First column is the spatial signature of the EEG in the source space, second column is temporal signature, third is the spectral signature and final column is the spatial signature of the EEG in the sensor space.

Figure 1
Image

Acknowledgements

This study is supported by the Bogazici University Scientific Research Fund under the project code 12XD2.

References

Bayram, A., Bayraktaroglu, Z., Karahan, E., Erdogan, B., Bilgic, B., Ozker, M., Kasikci, I., Duru, A.D., Ademoglu, A., Oztürk, C., Arikan, K., Tarhan, N., Demiralp, T. (2011). Simultaneous EEG/fMRI analysis of the resonance phenomena in steady-state visual evoked responses. Clin. EEG Neurosci. 42, 98-106.

Acar, E., Kolda, T.G., Dunlavy, D.M. (2011). All-at-once Optimization for Coupled Matrix and Tensor Factorizations. Proc. of KDD Workshop on Mining and Learning with Graphs.

Liu, W., Bailey, J., Leckie, C., Kotagiri, R. (2013). Mining Labelled Tensors by Discovering both their Common and Discriminative Subspaces. Proc.of the SIAM Conference on Data Mining.

Keywords: EEG-fMRI, multiway analysis, coupled tensor-matrix factorization, source imaging, DATA FUSION

Conference: Neuroinformatics 2013, Stockholm, Sweden, 27 Aug - 29 Aug, 2013.

Presentation Type: Poster

Topic: Neuroimaging

Citation: Karahan-Senvardar E, Duru A, Valdes-Sosa P and Ademoğlu A (2013). EEG-fMRI fusion on the cortical surface using Coupled Tensor-Matrix Factorization: A simulation study. Front. Neuroinform. Conference Abstract: Neuroinformatics 2013. doi: 10.3389/conf.fninf.2013.09.00088

Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.

The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.

Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.

For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.

Received: 08 Apr 2013; Published Online: 11 Jul 2013.

* Correspondence: Mrs. Esin Karahan-Senvardar, Bogazici University, Institute of Biomedical Engineering, Istanbul, 34684, Türkiye, esin.karahan@boun.edu.tr