EEG ‐To‐ fMRI Cross‐Modal Prediction via Hierarchical Temporal Attention Network: Toward Accessible Functional Neuroimaging
Concurrency and Computation: Practice and Experience · 2026-07 · 38 (14)
DOI: 10.1002/cpe.70860 ↗Kurz gefasst
Fachartikel in der Zeitschrift Concurrency and Computation: Practice and Experience, online erschienen am 14.07.2026. Verfasst von Lei Zhang und 4 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben.
Abstract (Originalsprache)
EEG and fMRI exhibit strong complementarity in temporal and spatial resolution; however, their substantial heterogeneity in signal generation mechanisms and temporal scales poses persistent challenges for EEG‐to‐fMRI cross‐modal prediction, particularly in terms of stability and interpretability. To address the limitations of existing approaches that rely heavily on voxel‐level high‐dimensional modeling or complex physiological assumptions, we propose a Hierarchical Temporal Attention Network (HTAN) for EEG‐to‐fMRI prediction, which takes ROI‐level BOLD signal generation as its primary modeling objective and explicitly characterizes the temporal dependencies between multiscale EEG neural dynamics and hemodynamic responses. HTAN constructs EEG temporal context inputs aligned with the fMRI sampling rate and incorporates hierarchical temporal encoding and attention mechanisms to effectively learn the contribution weights of EEG features across multiple temporal scales to ROI‐level BOLD responses. Experimental results on synchronized EEG–fMRI datasets demonstrate that HTAN achieves a group‐level mean scan‐wise Pearson correlation coefficient of 0.82 ± 0.02, with individual subject performance consistently distributed between 0.80 and 0.86, indicating strong cross‐subject generalization. By further projecting the predicted ROI‐level BOLD signals into anatomical voxel space to generate pseudo‐fMRI visualizations, spatial consistency analyses reveal stable and reproducible activation patterns within functionally relevant brain regions. It should be emphasized that this procedure serves solely as a spatial representation of ROI‐level predictions and does not constitute direct voxel‐wise fMRI reconstruction. These results demonstrate that, without resorting to voxel‐level high‐dimensional generation or explicit hemodynamic modeling, HTAN can robustly uncover the temporal mapping between EEG and fMRI BOLD signals through structured temporal modeling and function‐oriented ROI‐level analysis, providing a stable and neurobiologically interpretable paradigm for EEG–fMRI cross‐modal prediction.
Zitieren
APA
Zhang, L., Wang, J., Wang, H., Atzori, M., & Müller, H. (2026). EEG ‐To‐ fMRI Cross‐Modal Prediction via Hierarchical Temporal Attention Network: Toward Accessible Functional Neuroimaging. Concurrency and Computation: Practice and Experience, 38(14). https://doi.org/10.1002/cpe.70860
Vancouver
Zhang L, Wang J, Wang H, Atzori M, Müller H. EEG ‐To‐ fMRI Cross‐Modal Prediction via Hierarchical Temporal Attention Network: Toward Accessible Functional Neuroimaging. Concurrency and Computation: Practice and Experience. 2026;38(14). doi:10.1002/cpe.70860
BibTeX
@article{Zhang2026,
author = {Zhang, Lei and Wang, Jie and Wang, Hua and Atzori, Manfredo and Müller, Henning},
title = {EEG ‐To‐ fMRI Cross‐Modal Prediction via Hierarchical Temporal Attention Network: Toward Accessible Functional Neuroimaging},
journal = {Concurrency and Computation: Practice and Experience},
year = {2026},
volume = {38},
number = {14},
doi = {10.1002/cpe.70860}
}
EUVIMED