Mind the State: Towards Unified, Context-Aware EEG-to-fMRI Synthesis

ICLR 2026 Conference Submission16554 Authors

Keywords: EEG, fMRI, EEG-to-fMRI synthesis

TL;DR: We propose UniEFS, a unified model that reconstructs whole-brain fMRI from EEG with spatial priors and context-aware prompts, enabling robust, generalizable brain activity synthesis.

Abstract: Functional magnetic resonance imaging (fMRI) provides dynamic measurements of human brain activity at high spatial resolution and depth, but its use is constrained by high cost, limited accessibility, and strict acquisition requirements. Synthesizing fMRI data from more accessible, non-invasive modalities such as electroencephalography (EEG) offers a promising alternative, enabling inference of deep brain activity from low-cost scalp recordings in naturalistic settings. Despite recent progress, existing EEG-to-fMRI translation methods typically require training separate models for individual brain regions and offer limited consideration of subject-level variability in brain dynamics. In this study, we propose UniEFS, a unified EEG-to-fMRI synthesis model that enables full-brain fMRI reconstruction while accommodating datasets with varying demographic and physiological contexts within a single model. UniEFS leverages a pretrained fMRI decoder to embed rich spatial priors, as well as condition-aware prompt tokens that encode subject-level and experimental metadata to handle heterogeneous datasets. We extensively evaluate our model performance on eyes-closed resting-state data and demonstrate that it can reliably reconstruct temporally resolved whole-brain fMRI activity, with strong potential to generalize to task-based fMRI in a zero-shot setting.

Primary Area: applications to neuroscience & cognitive science

Submission Number: 16554