Simultaneous EEG-fMRI, a technique that records electrical brain activity and blood-flow-based brain imaging at the same time, has undergone a quiet revolution over the past several years. The combination gives researchers something neither tool offers alone: the millisecond timing of EEG married to the precise spatial maps of fMRI. Recent breakthroughs in deep learning artifact removal, ultra-high-field scanning at 7 Tesla, and AI models that can predict one signal from the other are pushing the technique from a niche research method toward broader clinical use in epilepsy surgery planning, consciousness detection, and psychiatric diagnosis.
Why Run Both at Once
EEG picks up the brain’s electrical signals through scalp electrodes. It captures changes on a millisecond timescale, which is fast enough to track individual neural events, but it cannot pinpoint exactly where in the brain those signals originate. fMRI, by contrast, maps blood-oxygen changes across the entire brain with millimeter-level precision, but it works on a timescale of seconds, far too slow to catch the rapid dynamics of thought and perception. Running both simultaneously lets researchers see when something happens (via EEG) and where it happens (via fMRI) in the same person, during the same task, at the same moment.
That pairing has revealed something fundamental about how the brain’s resting-state networks behave. EEG microstates, brief periods of roughly 100 milliseconds during which the pattern of electrical activity across the scalp stays stable, turn out to correspond to the large-scale resting-state networks that fMRI identifies. Researchers showed this by convolving the timing of EEG microstates with the sluggish blood-flow response and finding that four typical microstate patterns mapped onto four distinct fMRI networks.1PubMed. BOLD correlates of EEG topography reveal rapid resting-state network dynamics The implication is striking: these networks are switching on and off far more rapidly than fMRI alone would suggest, and only simultaneous EEG-fMRI can capture that speed.2PubMed Central. EEG microstate sequences in healthy humans at rest reveal scale-free dynamics
The Artifact Problem and How Deep Learning Is Changing It
The biggest practical barrier to simultaneous EEG-fMRI has always been artifact contamination. The MRI scanner’s rapidly switching magnetic gradients induce massive voltage spikes in the EEG electrodes, and the heartbeat creates a separate artifact called the ballistocardiogram (BCG), which smears rhythmic noise across EEG channels every time blood pulses through vessels near the scalp. Together, these artifacts can be orders of magnitude larger than the brain signals researchers are trying to measure.
Traditional correction methods, such as average artifact subtraction and independent component analysis, have been workhorses for over two decades. A systematic review of artifact reduction techniques published since 1998 found many available methods spanning hardware, model-based, and data-driven categories, but noted that few studies have adequately compared them head-to-head.3PubMed Central. Artifact Reduction in Simultaneous EEG-fMRI: A Systematic Review of Methods and Contemporary Usage Refinements like the adaptive optimal basis set approach have improved on older techniques, cutting average BCG residuals to about 5.5% compared with roughly 12.5% for standard average artifact subtraction and over 20% for independent component analysis.4Scientific Reports. Adaptive optimal basis set for BCG artifact removal in simultaneous EEG-fMRI
The real shift, though, is the arrival of deep learning. Recurrent neural networks trained on the nonlinear relationship between the heartbeat signal and the BCG artifact can now suppress the artifact while simultaneously improving the classification of task-relevant EEG features.5PubMed Central. Ballistocardiogram Artifact Reduction in Simultaneous EEG-fMRI Using Deep Learning Generative adversarial networks offer another route, learning to strip away the BCG artifact without needing additional reference signals or special hardware.6PubMed. Ballistocardiogram artifact removal in simultaneous EEG-fMRI using generative adversarial network A recent denoising autoencoder framework tackled both gradient and BCG artifacts at once, achieving a signal-to-noise ratio gain of about 15 decibels and strong structural similarity between cleaned and ground-truth signals, positioning it as a candidate for real-time correction.7arXiv. Deep Learning for Gradient and BCG Artifacts Removal in EEG During Simultaneous fMRI These AI-driven methods are not yet standard in every lab, but they are closing the gap between what simultaneous EEG-fMRI records and what researchers can actually use.
Ultra-High Field Scanning at 7 Tesla
Most simultaneous EEG-fMRI work has been done at 1.5 or 3 Tesla, the field strengths of standard clinical and research MRI scanners. Moving to 7 Tesla promises dramatically finer spatial resolution in the fMRI data, potentially down to sub-millimeter voxels that can distinguish individual cortical layers. But higher field strengths amplify every artifact problem and introduce new safety concerns, because the stronger magnetic field deposits more radiofrequency energy into the EEG hardware and surrounding tissue.
An optimized framework for 7T EEG-fMRI, developed on a clinical 7T system, recently demonstrated that sub-millimeter fMRI resolution could be achieved during simultaneous EEG recording without detectable safety issues or major practical constraints.8PubMed Central. An optimized framework for simultaneous EEG-fMRI at 7T enabling safe, high-quality human brain imaging with millisecond temporal resolution and sub-millimeter spatial resolution The key innovations included compact signal transmission between the EEG cap and amplifiers, reference sensors for artifact correction, and adapted leads compatible with dense radiofrequency receive arrays. Earlier foundational work at 7T had shown that shortening and bundling EEG cables reduced environmental noise by up to 84% in average power and 91% in the variability of noise across channels.9PubMed. Simultaneous EEG-fMRI at ultra-high field: artifact prevention and safety assessment
Researchers have also explored field strengths beyond 7T. A review covering acquisition up to 9.4 Tesla acknowledged that while safety concerns at ultra-high field have been effectively addressed, some loss in temporal signal-to-noise ratio from the EEG hardware remains acceptable for fMRI even with fairly high electrode densities.10NeuroImage. Simultaneous EEG–fMRI acquisition at low, high and ultra-high magnetic fields up to 9.4 T: Perspectives and challenges The practical upshot is that 7T EEG-fMRI is becoming feasible for routine research, which means future studies of cortical layer activity, columnar organization, and other fine-grained brain architecture can include the temporal dimension that EEG provides.
Epilepsy Surgery Planning
One of the most clinically mature applications of simultaneous EEG-fMRI is in epilepsy. For patients with drug-resistant seizures, surgery to remove the seizure focus can be life-changing, but only if the focus is correctly identified. Standard EEG localizes the electrical signature of seizures but struggles to see deep structures or map the entire network involved. fMRI alone cannot catch the brief interictal discharges that pinpoint the focus. Together, they offer a multidimensional view, mapping the blood-flow networks associated with specific epileptic events across the whole brain.11PubMed Central. Methods and utility of EEG-fMRI in epilepsy
A study of 37 patients demonstrated that the hemodynamic response to interictal epileptic discharges could localize the seizure-onset zone with practical accuracy: among studies where the primary activation cluster was explored with implanted electrodes, about 68% were concordant with the seizure-onset zone. When the statistical map showed a single dominant cluster, concordance could be predicted with greater than 90% confidence.12PubMed. The hemodynamic response to interictal epileptic discharges localizes the seizure-onset zone More recently, intracranial EEG-fMRI, where electrodes are implanted directly in the brain rather than placed on the scalp, has been used to evaluate the utility of these maps for predicting postsurgical outcomes.13Brain. Mapping interictal discharges using intracranial EEG-fMRI to predict postsurgical outcomes This invasive validation step is significant because it connects the noninvasive method to actual patient results, building a case for wider clinical adoption.
Detecting Consciousness After Brain Injury
Another area where EEG and fMRI are being deployed together, though not always simultaneously in the scanner, is cognitive motor dissociation: the situation in which a patient who appears unresponsive can nonetheless follow mental commands detected through brain imaging or EEG. fMRI might reveal that a patient asked to imagine playing tennis shows motor cortex activation, while EEG can catch command-following through changes in brain rhythms. Using both modalities increases the chance of detecting covert awareness that a single method might miss.
A report on clinical implementation of these techniques in an acute care hospital highlighted several practical lessons: the need for standardized local protocols covering patient selection, data acquisition, analysis, and interpretation, as well as careful ethical consideration of what it means to discover awareness in someone previously thought to be unconscious.14PubMed Central. Clinical Implementation of fMRI and EEG to Detect Cognitive Motor Dissociation: Lessons Learned in an Acute Care Hospital The field is still working out how to communicate test results to families and how to integrate these findings into care decisions, but the diagnostic potential is clear.
Predicting fMRI From EEG With AI
Perhaps the most futuristic advance is the effort to bypass the MRI scanner entirely by predicting fMRI-like brain maps from EEG data alone. A framework called NeuroBOLT uses a transformer architecture to learn mappings from raw EEG across temporal, spatial, and spectral dimensions to the corresponding fMRI blood-oxygen-level-dependent signals. It can reconstruct unseen resting-state fMRI signals from primary sensory areas, higher cognitive regions, and even deep subcortical structures, with the ability to generalize across different recording conditions and sites.15NeurIPS Proceedings. EEG fMRI: Latest Advances in Brain Research
If models like NeuroBOLT continue to improve, they could fundamentally change the cost and accessibility of brain imaging. EEG is portable, cheap, and can be used at the bedside, while fMRI requires a multi-million-dollar scanner and a cooperative, motionless patient. Being able to infer spatial brain information from a portable EEG headset, trained on paired EEG-fMRI data, would open brain mapping to settings where MRI is impractical: neonatal intensive care, rural clinics, longitudinal monitoring at home. The technology is still in its research phase, and the accuracy is not yet at clinical-grade levels for individual diagnostic decisions, but the trajectory is promising.
Neurovascular Coupling Is Not Uniform
A subtler but scientifically important advance involves what simultaneous EEG-fMRI has taught us about the relationship between neural activity and blood flow. The entire premise of fMRI rests on the assumption that when neurons fire more, local blood flow increases in a predictable way. Simultaneous EEG-fMRI has helped refine that assumption.
Research has shown that resting-state blood-flow fluctuations occurring after spontaneous neural “events” in the EEG closely resemble the blood-flow response to actual sensory stimulation, and this holds across different EEG frequency bands including theta, alpha, beta, and gamma.16PubMed. The resting-state neurovascular coupling relationship: rapid changes in spontaneous neural activity in the somatosensory cortex are associated with haemodynamic fluctuations that resemble stimulus-evoked haemodynamics But this coupling is not the same everywhere. Animal studies using combined electrophysiology and fMRI have found that the quantitative relationship between neural responses and fMRI responses in cortical regions is markedly different from the relationship in subcortical regions, indicating that coupling between neurons and blood flow is neither empirically nor mechanistically consistent across the brain.17PubMed. Regional differences in neurovascular coupling in rat brain as determined by fMRI and electrophysiology
This matters because many fMRI analyses assume a single, standard blood-flow response function for the entire brain. If that assumption is wrong in specific regions, the spatial maps derived from fMRI alone could be misleading. Simultaneous EEG-fMRI is one of the few tools that can expose where and how this assumption breaks down, which in turn helps calibrate fMRI results more carefully.
Gamma Oscillations and Decision Making
Simultaneous EEG-fMRI has also clarified the functional meaning of specific brain rhythms. Gamma oscillations, the fastest commonly studied EEG rhythms, have long been associated with perception and attention, but researchers debated whether different gamma frequencies reflected different brain processes or just noise. Using an ambiguous perception task where subjects either did or did not perceive a coherent object, a simultaneous EEG-fMRI study showed that a lower gamma band near 40 Hz was tightly linked to the decision-making network and particularly the anterior insula, while a higher gamma band around 60 Hz was connected to early visual processing regions.18PubMed Central. The dual facet of gamma oscillations: separate visual and decision making circuits as revealed by simultaneous EEG/fMRI Without the spatial information from fMRI, this distinction between gamma sub-bands would have been invisible in the EEG alone.
Sleep, the Thalamus, and Subcortical Mapping
Simultaneous EEG-fMRI is uniquely suited to studying the sleeping brain, because you can identify sleep stages and specific sleep events through EEG while watching the corresponding spatial patterns in fMRI. Work on sleep spindles, the brief bursts of rhythmic activity that occur during non-rapid-eye-movement sleep, revealed that slow spindles and fast spindles activate different brain regions. Both types involved the thalamus, anterior cingulate cortex, insula, and superior temporal gyri, but slow spindles were distinctly associated with the superior frontal gyrus, while fast spindles recruited sensorimotor cortex and the hippocampus.
The thalamus, a deep brain structure that acts as a relay station for sensory and cognitive information, has been particularly difficult to study with either method alone. A simultaneous EEG-fMRI study of working memory found that memory load influenced thalamic activity during the delay period between encoding and recall, with higher EEG source amplitudes in the thalamus for low versus high memory load, occurring within about 160 to 390 milliseconds of the delay period’s onset.19PubMed Central. A simultaneous EEG-fMRI study of thalamic load-dependent working memory delay period activity Capturing a subcortical timing difference at that resolution simply is not possible without combining both modalities.
Real-Time Neurofeedback
Neurofeedback, the idea that you can learn to regulate your own brain activity by watching it on a screen in real time, has been explored with both EEG and fMRI independently. EEG neurofeedback responds quickly but cannot target deep brain regions precisely. fMRI neurofeedback can target specific areas but updates too slowly to give moment-by-moment feedback. Simultaneous real-time EEG-fMRI neurofeedback combines the strengths of both, potentially letting someone learn to control activity in a specific brain region while getting feedback fast enough to make the learning intuitive.20PubMed Central. Simultaneous real-time EEG-fMRI neurofeedback: A systematic review The technology is still largely experimental, and the field has not yet converged on which feedback signals and training protocols work best, but the concept has attracted growing interest for potential applications in treating anxiety, chronic pain, and attention disorders.
Psychiatric Biomarkers
The search for objective biological markers of psychiatric conditions is one of the most challenging frontiers in brain research. A recent study used simultaneous EEG-fMRI to see whether EEG-derived dynamic connectivity features could predict an fMRI-based schizophrenia biomarker. The approach yielded significant classification accuracy in 19 out of 21 participants, and the biomarker showed specificity against a separate depression biomarker.21PubMed. Prediction of an fMRI-based schizophrenia biomarker from EEG using dynamic functional connectivity: a simultaneous EEG-fMRI study The idea here is similar to the NeuroBOLT approach but directed at a clinical question: can you eventually detect schizophrenia-related brain patterns using only EEG, once the link between EEG and fMRI signatures has been established through paired recording? If so, screening could move from expensive, scanner-based assessments to portable EEG setups.
Anesthesia and the Study of Consciousness
General anesthesia provides a controlled, reversible way to switch consciousness off and back on, making it a natural laboratory for studying what consciousness requires in the brain. EEG and fMRI studies of anesthesia have converged on a key finding: behavioral unresponsiveness under various anesthetic agents is consistently associated with a depression or disconnection of lateral frontoparietal networks, regions thought to be critical for awareness of the environment.22PubMed Central. Disconnecting Consciousness: Is There a Common Anesthetic End Point? Changes in connectivity patterns, network topology, and the timing of interactions between brain regions have also been documented during anesthesia.23Frontiers in Systems Neuroscience. General Anesthesia: A Probe to Explore Consciousness Because EEG captures the rapid dynamics of these transitions while fMRI reveals which specific networks are disconnecting, simultaneous recording has been instrumental in building current theories of how anesthetic drugs produce unconsciousness.
Aging and the Sensitivity of Combined Modalities
One of the quieter but potentially impactful findings from simultaneous EEG-fMRI research involves aging. A study comparing younger and older adults found that four out of 26 resting-state networks showed differences on both fMRI alone and combined EEG-fMRI analysis, but seven additional networks showed differences only when the EEG and fMRI signals were analyzed together, suggesting the combined approach is more sensitive to age-related neural changes than fMRI on its own. Activity within some of these combined networks was better predicted by neuropsychological performance measures than by age itself, hinting that the technique captures functionally meaningful brain changes rather than just structural decline.
New Electrode Materials
Hardware development continues to push the technique forward. One persistent annoyance is that conventional EEG electrodes contain metal, which interacts with the MRI’s magnetic field and can cause image distortion, heating risks, and extra artifacts. Researchers have developed MRI-compatible asymmetric bilayer hydrogel electrodes that use ionic conduction instead of metal conductors. These gel-based electrodes are also compatible with CT imaging, opening the possibility of seamless brain monitoring across different clinical imaging modalities without swapping electrode systems between scans.24Europe PMC. MRI and CT compatible asymmetric bilayer hydrogel electrodes for EEG-based brain activity monitoring The material science behind MRI-compatible electrodes does not get much popular attention, but it is one of the practical bottlenecks that determines whether simultaneous EEG-fMRI can scale beyond specialized research centers.
Animal Models and Cross-Species Validation
Much of what we understand about the neural basis of fMRI signals comes from animal experiments where invasive recordings can be made alongside imaging. Laboratory animals provide a platform for the kind of multi-dimensional, controlled investigation that is impossible in humans, and these experiments form the basis for interpreting the temporal correlations observed across the human brain.25PubMed Central. Contribution of animal models toward understanding resting state functional connectivity As simultaneous EEG-fMRI becomes more common in rodents and non-human primates, the feedback loop between animal and human research is tightening. Findings about neurovascular coupling variability in animals, for instance, are directly informing how researchers interpret human fMRI data, and human EEG-fMRI discoveries about resting-state dynamics are being validated and probed mechanistically in animal preparations. This cross-species conversation is one of the less visible but most scientifically valuable aspects of the field’s recent progress.