| Title:
|
EEG–fMRI Data Fusion via Generalized Coupled Matrix Tensor Factorization with Relaxed Similarity Constraint
|
| Author(s):
|
1
|
| Subject:
|
Advanced Techniques in Signal Acquisition and Image Processing
|
|
|
|
Presenter Author:
|
|
| Name:
|
Zahra Rabiei
|
|
|
|
Author:
|
|
| Name 1:
|
Zahra Rabiei - Corresponding - Presenter
|
| Affiliation 1:
|
Department of Electrical Engineering, AK.C.,Islamic Azad University, Aliabad Katoul, Iran
|
|
|
| Name 2:
|
|
| Affiliation 2:
|
|
|
|
| Name 3:
|
|
| Affiliation 3:
|
|
|
|
| Name 4:
|
|
| Affiliation 4:
|
|
|
|
| Name 5:
|
|
| Affiliation 5:
|
|
|
|
| Name 6:
|
|
| Affiliation 6:
|
|
|
|
| Name 7:
|
|
| Affiliation 7:
|
|
|
|
| Name 8:
|
|
| Affiliation 8:
|
|
|
|
| Name 9:
|
|
| Affiliation 9:
|
|
|
|
| Name 10:
|
|
| Affiliation 10:
|
|
|
|
Abstract
|
Original Research
|
|
introduction :
|
Simultaneous recording of electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) provides complementary spatiotemporal information about brain activity, but integrating these modalities is difficult because of their different biophysical origins and lack of direct temporal correspondence. Most fusion methods, such as advanced coupled matrix-tensor factorization (ACMTF), assume strict equality of shared components in the common mode, which is often too restrictive for real data. We developed a generalized coupled matrix-tensor factorization (GCMTF) approach that relaxes this equality constraint by introducing normalized mutual information (NMI)-based similarity penalty, allowing detection of both linear and nonlinear relationships between shared components
|
|
Methods:
|
In our study, EEG data from an auditory oddball paradigm were arranged as a third-order tensor (trials × frequency × channels) after bandpass filtering (1–40 Hz) and ICA-based artifact removal. Preprocessed fMRI data (motion correction, spatial smoothing, GLM-based activation mapping) were represented as a matrix (trials × voxels). Trial was the common mode, with fMRI trials convolved with a canonical hemodynamic response function (HRF). Unlike ACMTF, which enforces identical shared trial factors, the proposed GCMTF incorporated an NMI-based similarity term in the cost function as an adaptive penalty. Optimization was performed using gradient-based algorithms with histogram-based NMI estimation. The model was evaluated against ACMTF and related studies.
|
|
Results:
|
On simulated datasets with varying linear and nonlinear relationships between modalities, GCMTF improved average match score by 23.46% over ACMTF across different noise levels. In real simultaneous EEG-fMRI data from 17 healthy subjects performing an auditory oddball task, three shared components with frequency responses in theta and alpha bands were identified. The method also revealed more active brain regions corresponding to target stimuli compared to ACMTF and similar approaches.
|
|
Conclusion:
|
The proposed GCMTF method effectively captures complex linear and nonlinear dependencies between EEG and fMRI data that are missed by strict equality-based fusion approaches. By enabling more accurate identification of shared and modality-specific components, this framework provides deeper insight into neural processes underlying cognitive tasks such as auditory attention and offers a robust tool for multimodal brain imaging studies.
|
| Attachment:
|
|
| Keywords:
|
Data fusion, Coupled matrix tensor factorization, EEG-fMRI, Normalized mutual information, Auditory oddball.
|
| Key message:
|
|
| Do you want your abstract to be indexed, if this is selected as one of the top 100
articles?
|
