Abstract
Simultaneous EEG-fMRI combines the temporal resolution of electrophysiology with the spatial coverage of blood-oxygen-level-dependent imaging, but these benefits come with substantial methodological constraints. We examine simultaneous EEG-fMRI as a linked acquisition-to-inference workflow, with the two modalities treated as interdependent during preprocessing and analysis. We consider MRI-compatible acquisition and safety, temporal synchronization, gradient and ballistocardiographic artifact correction, EEG and fMRI preprocessing, quantitative quality assessment, connectivity analysis, and multimodal fusion. We focus on the assumptions and failure modes of each approach, preservation of neural information, and evidence for generalization beyond the setting in which each method was developed. The available literature does not support a single artifact-correction strategy for all recordings. Template methods remain effective when artifact timing and morphology are stable, whereas basis-set, adaptive, reference-based, and learning-based approaches may be more appropriate when contamination varies over time or real-time operation is required. Because preprocessing can alter spectral estimates, connectivity, and network organization, residual artifact amplitude alone is an inadequate measure of success. We propose a decision-oriented framework in which acquisition conditions, artifact structure, analysis target, and validation level are considered together when selecting an EEG-fMRI pipeline.