RESEARCH DOSSIER
EEG, MEG, and Neural Temporal Dynamics
Abstract. This fundamental research monograph investigates how electroencephalography (EEG) and magnetoencephalography (MEG) measure the temporal dynamics of neural population activity and how those measurements inform psychological explanations. It traces the historical development of human electrophysiology and neuromagnetic recording, examines neural current generators and inverse models, and compares event-related potentials, induced oscillations, time-frequency decompositions, aperiodic spectra, phase relations, source imaging, and connectivity inference. Methodological chapters analyze acquisition quality, filtering, artifact identification, single-trial variation, multiplicity correction, reproducibility, and data standards. Research applications cover perception, attention, memory, action, language, development, and clinical measurement; a contemporary synthesis evaluates optically pumped wearable magnetometry, combined EEG/MEG recording, and repeat-test reliability evidence through 2025. The evidence supports careful separation of observable neural timing, statistical dependence, candidate physiological mechanisms, and psychological construct attribution. Major open problems concern transient versus rhythmic dynamics, source leakage, valid causal interpretation, test-retest stability, cross-population generalization, and naturalistic acquisition. This is a critical scholarly research synthesis, not a protocol-registered systematic review or a meta-analysis.
Keywords: electroencephalography; magnetoencephalography; EEG; MEG; neural temporal dynamics; event-related potentials; neural oscillations; time-frequency analysis; aperiodic activity; phase synchronization; source localization; inverse problem; neural connectivity; cognitive neuroscience; wearable magnetometry; research reproducibility
The Research Object: Neural Events, Time, and Psychological Explanation
Electroencephalography (EEG) and magnetoencephalography (MEG) register the evolving electromagnetic consequences of coordinated neural population activity. Their scientific object is the temporal organization of measurable brain signals in relation to experimental events, ongoing states, and behavior. Temporal dynamics includes response onset, duration, amplitude modulation, phase structure, recurring rhythms, transient bursts, coupling between populations, and the variability of those properties across trials. The psychological questions concern perception, attention, working memory, action preparation, learning, language, and the organization of behavior as unfolding processes. Because recordings mix signals from neural generators, interpretation requires a biophysical forward model and an independently specified behavioral construct Cohen (2017) Baillet (2017).
The central distinction is between what an instrument observes and what a psychological explanation attributes. A voltage deflection at an electrode is an electrical potential difference; a magnetic-field fluctuation reflects current flow. A labeled peak does not itself constitute a recognized object, an inhibition operation, or a decision. Those are hypotheses about processes that require task manipulations, behavioral evidence, and convergence with other measurements. An observed temporal order between two signal features also does not by itself prove that the earlier source caused the later process. This monograph treats the inferential route from physical measurement to temporally resolved cognition as its primary methodological problem Nunez & Srinivasan (2006) Gross et al. (2013).
The research synthesis draws on historical original reports, experimental studies, electrophysiological measurement theory, quantitative signal-processing methods, reproducibility guidelines, and recent instrumentation advances. It compares event-related averages with single-trial dynamics, oscillatory with aperiodic activity, sensors with reconstructed sources, and predictive associations with causal mechanisms. The review is a critical narrative integration rather than a protocol-registered systematic review; no exhaustive database coverage, PRISMA flow, or pooled effect estimate is claimed. Its boundary is noninvasive human EEG and MEG, with invasive recordings and perturbations considered where they help identify sources or test a mechanism. Neighboring dossiers address cellular excitability, network connectomics, functional MRI, and stimulation as distinct principal objects.
The Historical Emergence of Electrical Brain Recording
The modern human EEG record is conventionally traced to Hans Berger's 1929 publication on the human electroencephalogram. Berger established that potential fluctuations measurable at the scalp could be studied as ongoing physiological phenomena, laying the basis for subsequent attention to spontaneous rhythms and state differences Berger (1929). Early descriptions preceded modern amplifier, electrode, and computational standards; their historical significance lies in rendering the changing electrical activity of the living human brain an experimentally accessible object. A modern signal cannot be interpreted simply by reproducing period terminology, because measurement circuitry, recording reference, artifact control, and spectral estimation changed radically during the development of the field.
The significance of scalp electrical measurement expanded as investigators shifted from describing resting rhythms toward time-locking recordings to controlled sensory, cognitive, and motor events. Averaging repeated stimulus trials improved extraction of weak responses amid variable background activity, while laboratory psychophysiology developed standardized tasks and response components. The subsequent literature on selective attention used event-related potentials to ask when sensory selection influences processing, and whether attentional operations affect early and late phases differently Luck et al. (2000). The historical transformation is methodological as well as conceptual: recording cortical activity became a way to test a sequence of experimentally specified processing claims rather than a simple inventory of brain-wave types.
Digital acquisition enabled storage of continuous multichannel data and single trials, opening the possibility of revisiting assumptions concealed by trial averages. Independent component analysis, spectral decomposition, source modeling, and increasingly open processing software made methodological choices both powerful and consequential. Delorme and Makeig's EEGLAB platform illustrates the transition from conventional averages to integrated analyses of single-trial variability and signal components Delorme & Makeig (2004). The current research problem retains the historical commitment to observing electrical activity while requiring a transparent account of how acquisition and computation construct each reported temporal feature.
The Development of Magnetoencephalography
MEG emerged from the ability to detect extremely weak magnetic fields associated with neuronal currents, with superconducting quantum interference devices permitting increasingly practical multichannel recording. The foundational account by Hämäläinen and colleagues describes instrumental principles, the electromagnetic forward problem, and early experiments tracking spontaneous and evoked brain activity Hämäläinen et al. (1993). Magnetic recording added a field geometry with different sensitivity to neural orientations and surrounding tissues. It opened additional opportunities for observing fast cortical activity and reconstructing spatially distributed sources without replacing EEG or providing a direct image of individual neurons.
Later MEG research emphasized sequences of cortical events at millisecond timescales and the relation between localized activity and distributed coordination. Hari and colleagues surveyed sensory, motor, cognitive, and social applications to explain how neuromagnetic recordings can characterize activity sequences Hari et al. (2010). Baillet clarified both the distinctive advantages of MEG source imaging and the limits imposed by noise, head geometry, inverse assumptions, and signal interpretation Baillet (2017). These developments support a combined research strategy in which electrical and magnetic measurements constrain shared but partially distinct aspects of the same neural current distributions.
The comparison of historical MEG instruments with newer wearable approaches also changes experimental possibilities. Traditional cryogenic arrays constrain the position of the head relative to fixed sensors; optical magnetometers have enabled development of wearable systems that tolerate some naturalistic movement under specialized magnetic-field control Boto et al. (2018) Tierney et al. (2019). These are engineering demonstrations with clearly defined acquisition environments, and their existence does not establish effortless brain recording in ordinary uncontrolled settings. They introduce a wider set of experimental designs for developmental and action research while retaining the need for validation of motion, interference, sensor calibration, and source reconstruction.
Neural Current Generators and the Electromagnetic Forward Problem
Scalp potentials and extracranial magnetic fields originate largely from spatially organized electrical currents in neural tissue, especially synchronized postsynaptic activity in suitably aligned cortical populations. The contributions observable at a distant sensor depend on geometry, temporal coherence, tissue conductivities, and the distribution and orientation of the active populations Nunez & Srinivasan (2006) Cohen (2017). A scalp EEG trace therefore aggregates information over spatially extended tissue and cannot isolate a microcircuit by inspection. The electrical field and magnetic field are generated by related currents, but they are shaped differently by the anatomy and measurement system.
Forward modeling predicts what a specified candidate source configuration would produce at sensors. Realistic head models may use anatomical MRI to represent the scalp, skull, cerebrospinal fluid, and cortical surfaces; computational approaches convert source currents into expected voltages or magnetic fields. Solving the forward problem requires known or estimated geometry and physical assumptions. The inverse problem moves in the opposite direction, seeking neural sources that could have generated the observed data. Because many source patterns can fit the same external measurements, the inverse mapping is nonunique and its solution depends on constraints Hämäläinen et al. (1993) Gramfort et al. (2014).
Interpretation should therefore proceed from instrument physics through signal modeling to psychological claims. Source estimates are conditional solutions to an assumed generative model, and their spatial smoothness, depth sensitivity, and timing depend on regularization and noise models. EEG and MEG have complementary sensitivity patterns, so convergent evidence can help distinguish candidate generators. Still, convergent localizations do not automatically identify a psychological operation: the same population can participate in distinct tasks and a process can involve multiple regions. This distinction motivates experimental contrasts that test mechanistic alternatives rather than using a sensor map as direct representation of a mental faculty Baillet (2017) Gross et al. (2013).
Electroencephalography: Acquisition, Referencing, and Measurement Validity
EEG records voltage differences between electrodes and a chosen reference, so any reported channel waveform is defined by its referencing scheme. Electrode position, contact quality, amplifier properties, line-frequency contamination, eye movements, muscle contractions, and cable movement all affect measurement. High-density montages improve spatial sampling but do not remove the inverse problem. The meaning of a positive or negative deflection depends on reference and source orientation, making polarity alone a poor indicator of excitation, inhibition, or the sign of a psychological process Nunez & Srinivasan (2006) Keil et al. (2014).
A rigorously specified EEG experiment documents the task, stimulus timing, response devices, number and placement of channels, recording reference, sampling rate, online filtering, electrode impedances or comparable acquisition checks, rejection decisions, and preprocessing pipeline. These are scientific variables because each can change which dynamics appear measurable. Artifact procedures should identify plausible ocular, cardiac, muscle, and environmental signals while preserving task-related neural variation. Removing a component merely because it resembles a stereotyped artifact can erase a genuine experimental effect if the relevant physiological and behavioral events covary Delorme & Makeig (2004) Keil et al. (2014).
Measurement validity rests on whether the acquisition system reliably distinguishes the neural feature relevant to the hypothesis. A laboratory can recover a reproducible ERP while still failing to measure a distinct source or single-trial onset. Multiple preprocessing settings may yield similar waveforms but different amplitude or latency estimates. Researchers should retain continuous recordings, event logs, channel information, bad-channel decisions, and sufficient provenance to reproduce transformations. The EEG-BIDS specification provides an explicit route to share these data and metadata in interoperable structure Pernet et al. (2019). This makes the technical history of the signal auditable by later investigators.
Magnetoencephalography: Sensors, Interference, and Complementarity
MEG measures magnetic fields through sensors positioned outside the head. Traditional sensor arrays have different channel geometries, including magnetometers and gradiometers, and consequently different responses to external interference and intracranial sources. Magnetic signals from the brain are extremely weak relative to many environmental sources, making shielding, calibration, head localization, interference rejection, and movement tracking central to valid acquisition. Importantly, fields measured by MEG are not a second form of electrical activity independent of physiology: they are another observable consequence of neuronal current distributions Hämäläinen et al. (1993) Gross et al. (2013).
Signal-space separation exploits spatial patterns constrained by the magnetic field equations to distinguish signal subspaces associated with internal and external sources. Taulu and Simola introduced a spatiotemporal strategy that combines this geometry with temporal information to reject certain external and nearby artifacts Taulu & Simola (2006). Such processing increases usability but can introduce dependencies on sensor calibration, geometry, reconstruction choices, and assumptions about the spatial extent of sources. A trustworthy MEG report includes hardware characteristics, acquisition environment, sensor treatment, head movement, processing parameters, and whether comparisons across sessions were affected by differing signal-space projections Gross et al. (2013).
EEG and MEG differ in their sensitivity to source orientation, tissue conductivity, measurement noise, and deep generators. Their complementarity is most scientifically productive when the joint model is specified rather than when apparent similarity is counted as simple replication. The emerging optically pumped magnetometer platforms move sensors closer to the head and permit new forms of mobility, yet their experimental claims must be evaluated against contemporary noise-control requirements and the scope of each validation study Boto et al. (2018) Tierney et al. (2019). The practical conclusion is to select modalities from the required source geometry, population, behavior, and experimental constraints.
Evoked, Induced, and Phase-Reset Accounts
Evoked responses and induced oscillatory changes are distinguished analytically by their relation to event timing. In the usual formulation, an evoked signal remains sufficiently phase-aligned across trials to contribute to the time-domain average, whereas induced power modulation may occur without stable phase alignment. This classification describes measurable properties, and interpretation becomes more complex when a stimulus changes both amplitude and phase or produces transient responses with trial-to-trial latency variability. Tallon-Baudry and colleagues showed that phase-locked and non-phase-locked high-frequency responses could be investigated separately in visual stimulation experiments Tallon-Baudry et al. (1996).
One explanation of an event-related waveform is the addition of a transient response to ongoing oscillations. An alternative hypothesis is that the event reorganizes the phase of ongoing activity, yielding a repeatable average without the same simple additive generator. Makeig and colleagues analyzed dynamic sources of visual evoked responses and helped demonstrate why phase and independent components matter in interpreting averages Makeig et al. (2002). Both explanations can fit selected observations, and combinations are possible. The scientific test is to identify discriminating predictions in single-trial phase, baseline amplitude, source activity, and stimulus dependence rather than accepting an averaging artifact as a complete physiological mechanism.
Estimating phase across short windows can itself introduce misleading structure because a phase estimate presupposes an adequate oscillatory signal in the chosen band. Event-related changes in broad power or waveform shape can alter measured phase consistency even when a narrowband oscillator is not clearly identified. Filtering and wavelet selection affect these apparent dynamics, particularly near transient onset. Accordingly, phase-reset claims should include control simulations, alternative analytic settings, and a check that the proposed ongoing rhythm is measurable before the event. The value of the evoked-versus-induced distinction lies in formulating testable dynamical accounts rather than asserting that one classification directly reveals how perception occurs Cohen (2017) Widmann et al. (2015).
Spectral Decomposition and Neural Oscillatory Activity
The frequency spectrum of EEG or MEG summarizes how signal variance is distributed across rates of fluctuation. Narrowband peaks may reflect rhythmic population processes, while broadband activity can arise through nonrhythmic changes, transient events, noise, or mixtures of generators. A spectral peak does not imply that neurons across the sampled tissue fire at that rate, and equal band-power measurements can arise from different combinations of amplitude, burst duration, and spatial source mixing. Physiological interpretation therefore requires inspection of temporal structure and appropriate statistical baselines, as emphasized by modern analyses of neural power spectra Donoghue et al. (2020) Cohen (2017).
Frequency bands such as delta, theta, alpha, beta, and gamma are useful conventions that support comparisons across experiments, yet boundaries vary with population, recording site, task, and analysis. Individual dominant frequencies can shift with age or state, while source-specific rhythms may occupy overlapping conventional ranges. Differences in band power can reflect changed rhythmic amplitude, frequency shift, altered aperiodic background, or different artifact composition. A defensible report specifies actual frequency limits and the reason for them, describes spectral resolution, and considers individualized peaks when appropriate Klimesch (2012) Donoghue et al. (2020).
Spectral estimation methods include Fourier analysis of stationary windows, multitaper approaches designed to manage variance and spectral leakage, and time-frequency techniques for changing signals. Each introduces a tradeoff among frequency precision, time localization, and uncertainty. Oscillatory interpretations should not be inferred from visually persuasive color maps alone; they require hypotheses about coherent rhythmic dynamics and evidence that the selected processing pipeline can detect them under simulated signal conditions. The foundational contribution of spectral methods is a quantitative vocabulary for neural timescales, not a fixed correspondence between frequency label and psychological faculty Gross et al. (2013) Oostenveld et al. (2011).
Periodic and Aperiodic Components of Neural Spectra
Aperiodic spectral structure commonly approximated by a power-law-like background can strongly influence the magnitude of conventional band-power estimates. The exponent and offset of that background are descriptive features whose physiological causes may involve multiple processes. Donoghue and colleagues developed methods to parameterize periodic peaks together with an aperiodic component, demonstrating why changes in raw power cannot automatically be interpreted as changes in oscillation strength Donoghue et al. (2020). The distinction matters across resting conditions, task contrasts, development, and comparisons between patient groups in which global spectral shape may differ.
Parameter separation is itself model-dependent. The frequency interval chosen for fitting, spectral knees, peak widths, harmonic structure, filtering, and measurement noise can change estimated background parameters. A spectrum with bursts, transients, and several weak broad peaks may not meet the assumptions of a smooth background plus independent narrowband oscillations. The fitting procedure should therefore report goodness of fit, selected range, model specification, and sensitivity analyses. Treating an exponent as a direct measure of excitation-inhibition balance or any single biological parameter exceeds what spectral shape alone establishes Donoghue et al. (2020) Nunez & Srinivasan (2006).
This decomposition encourages more precise questions. A change in alpha power can be separated into a shift of the alpha peak and a changed broadband offset; a purported gamma enhancement can be examined for muscle contamination and aperiodic change. In either case the result remains a measurement-level fact until neural generators and behavioral functions are independently tested. Inference is strongest when raw spectra, fitted components, time-domain signals, and task conditions are available together, allowing other investigators to revisit the chosen decomposition Pernet et al. (2019) Jas et al. (2018).
Time–Frequency Estimation and Transient Bursts
Classical Fourier representations summarize frequency content over an analysis interval, but neural states change in response to events and during ongoing action. Short-time Fourier transforms, wavelets, and related methods localize frequency-dependent changes to approximate temporal windows. Resolution is bounded by the number of cycles required to estimate a frequency: broad temporal windows resolve nearby frequencies better but smear rapid changes, whereas narrow windows yield more uncertain frequency separation. Interpretation must distinguish the recorded sample interval from the effective temporal precision of the estimator Cohen (2017) Oostenveld et al. (2011).
An average time-frequency map can make irregular short bursts appear to be sustained oscillatory activation. Changes in burst probability, duration, instantaneous frequency, phase consistency, and peak amplitude can all raise average band power. Trial-resolved estimates can distinguish these alternatives when sufficient signal-to-noise ratio and robust event definitions are available. Such analyses also expose the weakness of defining a physiological burst by one arbitrary amplitude threshold: changing the threshold can change the measured rate and apparent task effect. Sensitivity to event selection and estimator bandwidth should be reported as part of the scientific result Donoghue et al. (2020) Makeig et al. (2002).
Time-frequency statistics have many correlated comparisons across time bins, frequencies, sensors, and participants, making informal selection of apparent peaks unreliable. Baseline periods must be physiologically appropriate; a pre-event interval may itself contain anticipatory changes. A response-locked analysis may reveal processes hidden by stimulus-locked averaging, but its relation to behavior can also reflect confounding from variable trial length. Statistical inference should therefore follow the sampling and randomization design, and cluster corrections should be interpreted at the level they actually test Maris & Oostenveld (2007) Sassenhagen & Draschkow (2019).
Alpha Oscillations: Attention and Functional Gating
Human alpha-band activity is prominent across many recording contexts, but its spatial distribution and psychological associations change with task demand. Work on alpha oscillations and attentional control supports models in which selected forms of rhythmic population activity regulate effective access to sensory representations and stored information. Jensen and Mazaheri argued for an inhibitory gating account, while Klimesch synthesized relations among alpha, attention, and controlled retrieval Jensen & Mazaheri (2010) Klimesch (2012). These are mechanistic models requiring experimental validation in particular circuits; lower or higher alpha power cannot be given one universal mental meaning.
In spatial cueing experiments, hemisphere-specific alpha changes may occur during anticipation, alongside changes in subsequent sensory responses and behavioral selection. Such relationships are informative because they link a specific experimental manipulation to a temporal signal and a measurable performance outcome. However, participants may vary in preferred alpha frequency, strategy, state, and artifact profile. Spectral estimates can also aggregate sources with different functional roles. Distinguishing proactive gating from a consequence of attentional effort requires manipulation, timing analyses, and independent behavioral predictions Brignani et al. (2009) Jensen & Mazaheri (2010).
Alpha research demonstrates the central principle of this dossier: a temporal signal can constrain competing descriptions of a psychological operation without being identical to that operation. Models of functional gating become stronger when alpha changes predict trial-specific selection, are spatially consistent with the relevant sensory pathways, and survive different preprocessing and statistical choices. Where results differ across tasks, the scientific task is to explain their dependence on stimulus, representation, and behavioral goal rather than to declare a single invariant interpretation of a familiar rhythm Klimesch (2012) Keil et al. (2014).
Theta, Beta, Gamma, and the Limits of Functional Specificity
Theta-band dynamics are frequently investigated in memory, cognitive control, and navigation, whereas beta activity is often examined in sensorimotor preparation and ongoing motor states. Gamma-frequency activity has been studied in sensory integration and other rapid coordination processes. These associations are useful entry points for research, but functional interpretation depends on source, timing, task, and behavioral endpoint. Frequency alone cannot uniquely identify a computational operation, because different neural systems can express similar oscillation frequencies and the same population can change frequency content across conditions Cohen (2017) Pfurtscheller & Lopes da Silva (1999).
Sensorimotor studies provide an instructive example. Event-related changes in beta-band power before and after movement have been repeatedly demonstrated in EEG and MEG research, and can be investigated alongside reaction time, preparation, inhibition, and sensory feedback. A suppression of power does not immediately reveal a population's firing rate, and a rebound following movement need not indicate a simple reversal of the preparatory computation. Separating efferent commands from reafferent responses requires experimental manipulation of movement conditions and attention, with source estimates or complementary recordings used cautiously Pfurtscheller & Lopes da Silva (1999) Baillet (2017).
High-frequency analysis raises additional challenges because facial and neck muscle activity can generate considerable spectral power in ranges frequently described as gamma. The distinction between stimulus-locked and induced activity is equally important when a visual or auditory response contains transient broadband energy. Research on visual responses has demonstrated separable rhythmic patterns under controlled stimulation, but the inference to a psychological function still needs alternative tasks and behavioral checks Tallon-Baudry et al. (1996). Frequency ranges should therefore be treated as descriptors of measured signals and candidate dynamics, while specialized functions are identified through convergent tests.
Phase Synchronization and Network Communication
Neural synchrony can describe correlated timing within or between neural populations, but its measured expression depends on the chosen statistic. Coherence combines amplitude and phase relations; phase-locking values emphasize the distribution of phase differences. Varela and colleagues organized a research program relating transient phase synchrony to the integration of distributed brain activity Varela et al. (2001). Fries proposed communication through coherence as one possible account of how relative oscillatory timing influences effective communication between neural groups Fries (2005). These are testable mechanistic hypotheses about coordinated signaling, not direct observations of message transfer from a synchronization score.
At the sensor level, a shared source can project to multiple electrodes or magnetometers and create apparent zero-lag synchrony without two independent neural generators interacting. Instantaneous field spread and source leakage persist even after mathematical source reconstruction. Nolte and colleagues proposed the imaginary part of coherency to identify a subset of delayed coupling less susceptible to certain instantaneous mixing effects Nolte et al. (2004). This helps with a particular confound but cannot establish true causality or recover all physiological interactions; real coupled sources may also synchronize near zero lag. Multiple complementary measures and realistic simulations are therefore necessary.
Statistical connectivity must be interpreted against the sample-size dependence of its estimator and against bias caused by power differences, measurement noise, and signal-to-noise imbalance. Vinck and colleagues introduced a weighted phase-lag index and associated analyses to reduce several of these problems Vinck et al. (2011). Haufe and colleagues demonstrated through simulations that different connectivity measures can behave differently under mixed-source conditions Haufe et al. (2013). Consequently, network diagrams inferred from EEG/MEG cannot be treated as literal maps of directed information exchange unless the experiment supplies evidence beyond pairwise statistical dependence.
Volume Conduction, Source Leakage, and Spurious Connectivity
Volume conduction in EEG and field spread in MEG mean that sensor measurements contain overlapping mixtures of activity from multiple neural populations. Even when sources are spatially distant, their projected sensor patterns may overlap; conversely, a single cortical generator may influence many channels. Correlated signals can therefore arise without an interaction among the putative locations represented by the channels. This physical fact is foundational for interpreting scalp network analyses, coherence, and phase synchrony Nunez & Srinivasan (2006) Hämäläinen et al. (1993).
Inverse reconstruction does not eliminate the problem because each estimated source has a point-spread and cross-talk function determined by sensor geometry, head model, noise, and regularization. Spatial leakage correction can reduce certain false dependencies but may alter true interactions or create novel distortions. Simulation studies that place known interacting and noninteracting sources into realistic forward models can reveal which network claims are identifiable for a particular measurement system. Methods based on the imaginary component or lagged relations trade resistance to instantaneous mixing against reduced sensitivity to true near-zero-lag coupling Nolte et al. (2004) Haufe et al. (2013).
A robust network study reports the assumed source model, spatial resolution, leakage diagnostics, statistical estimator, threshold decisions, and stability across plausible settings. It also asks whether the observed relation survives controls for common task events, power changes, and shared physiological artifacts. If functional communication is the hypothesis, temporal precedence and selective experimental perturbation provide additional discriminating evidence. Connectivity scores remain valuable measurements of covariance structure, while mechanistic claims require a chain of observations showing why independent neural populations must interact in the proposed manner Fries (2005) Vinck et al. (2011).
Source Reconstruction: Dipoles, Minimum Norm, and Beamforming
Source reconstruction uses an electromagnetic forward model to infer candidate distributions of cortical activity from external measurements. Equivalent-current dipole models describe a limited number of compact generators, whereas distributed models estimate activity across a larger source space. Minimum-norm approaches use regularization to select one solution among many compatible alternatives; beamformers use spatial filtering to estimate source contributions while suppressing signals from other locations. Each approach embeds assumptions about generator geometry, covariance, and noise, and their solutions answer different scientific questions Hämäläinen et al. (1993) Gramfort et al. (2014).
Beamforming can be advantageous when the question concerns oscillatory power changes from separable sources, but correlated generators and imprecise covariance estimation can impair localization. The linearly constrained minimum variance framework of Van Veen and colleagues formalized one influential spatial-filtering approach Van Veen et al. (1997). Minimum-norm reconstructions can reveal distributed activation patterns, but spatial smoothing and depth bias influence which areas appear active. Individual structural MRI and accurate sensor-to-head co-registration improve forward modeling yet do not convert a conditional source estimate into anatomically certain evidence.
Spatial localization should be evaluated through simulations, resolution matrices, known evoked responses, and comparisons with convergent imaging. Dynamic statistical parametric mapping provides one example of integrating MEG with MRI-based structural and functional information, illustrating both the promise and modeling demands of multimodal source imaging Dale et al. (2000). Open implementations such as MNE allow investigators to inspect and reproduce model choices Gramfort et al. (2014). The central epistemic rule is that estimated sources are model-dependent representations of measured data, which can constrain—but cannot independently settle—the anatomical origin or psychological interpretation of a dynamical process.
Multimodal Convergence and Temporal–Spatial Integration
EEG and MEG contribute high temporal resolution, while structural MRI, functional MRI, and invasive physiology can provide different forms of spatial and physiological constraint. Combining modalities can strengthen interpretation when each supplies an independent measurement of the same proposed event or mechanism. The dynamic mapping work by Dale and colleagues exemplifies integrating hemodynamic and electromagnetic information in source analysis Dale et al. (2000). Yet hemodynamic responses unfold on substantially slower timescales than electrophysiological measurements, so coincidence in anatomical label does not establish identity of underlying events. Models must specify the mapping across timescales.
Convergence can mean several scientifically distinct achievements: agreement about a broad region, reproducibility of a task-related temporal ordering, agreement between a neural signal and a behavioral manipulation, or improved predictive performance in independent participants. These should be reported separately. The value of an additional modality is greatest when it rules out an alternative explanation that one modality alone could not distinguish. EEG and MEG are particularly useful together because their sensitivity profiles differ and because their simultaneous or sequential recording can test the robustness of a proposed source model Baillet (2017) Gross et al. (2013).
More sensors or modalities do not automatically identify more mechanisms. Joint analyses can compound sources of error, including co-registration uncertainty, preprocessing dependencies, population differences, and analytic flexibility. Researchers should specify whether integration is constrained, data-driven, hypothesis-tested, or exploratory, and whether model performance is evaluated independently. The strongest multimodal explanation accounts for measured signal formation, temporal dynamics, source geometry, and task-specific behavioral predictions in one coherent causal framework Gramfort et al. (2014) Cohen (2017).
Preprocessing, Filtering, and Artifact Modeling
Preprocessing organizes acquired signals into an analysis-ready representation, but each transformation can alter the amplitude, timing, spectral shape, or spatial distribution of the reported effect. High-pass filtering may introduce systematic biases into early event-related responses, particularly when large slow components are present. Acunzo and colleagues demonstrated such distortions in ERP and ERF analysis, establishing the need to evaluate filters as part of the scientific model rather than treating them as harmless cleaning steps Acunzo et al. (2012). Widmann and colleagues developed practical principles for electrophysiological digital filter design that emphasize filter response, edge effects, and transparent specification Widmann et al. (2015).
Eye movements, blinks, cardiac signals, motion, and muscle activity can appear in both EEG and MEG. Independent component analysis can help identify mixtures with separable spatial and temporal characteristics, but component classification involves uncertainty and may remove neural activity correlated with the artifact. A report should identify automated and manual decisions, demonstrate their influence on primary contrasts, and retain enough information to allow a second analyst to test alternatives. The EEGLAB framework provided a widely adopted architecture for component-based inspection and single-trial analysis Delorme & Makeig (2004).
Preprocessing should be tied to the estimand: a pipeline appropriate for measuring a low-frequency slow potential may differ from one optimized for high-frequency oscillatory bursts. Resampling, baseline correction, channel interpolation, reference conversion, and artifact exclusion affect the distribution of subsequent statistics. Blinded and prespecified procedures reduce opportunities for selecting the pipeline that creates the desired contrast. Publishing original acquisition settings together with processing code and sensitivity analyses enables investigators to decide whether the substantive claim survives plausible choices Keil et al. (2014) Jas et al. (2018).
Single-Trial Variability and Ongoing Cortical State
Neural responses differ from trial to trial even when the stimulus and behavioral task are nominally repeated. This variability can originate in changing network state, attention, arousal, sensory history, and experimental noise, as well as differences in the timing and magnitude of the actual response. Arieli and colleagues combined population observations to demonstrate how ongoing activity contributes to variability in evoked responses, a foundational result in state-dependent neurophysiology Arieli et al. (1996). The lesson is that the condition average is an informative statistical estimate but only a partial description of the dynamics produced on individual occasions.
Single-trial approaches ask whether behavior is better predicted by phase, amplitude, latency, state variables, or the interaction between ongoing activity and the external event. Those relationships need designs that distinguish neural variability from changes in stimulus delivery, muscular output, and artifact contamination. A trial with a delayed neural component may be averaged into a low-amplitude response even if the neural event is strong in individual trials. The single-trial component framework represented in EEGLAB and related tools helps measure such heterogeneity, but recovered components remain model-dependent mixtures Delorme & Makeig (2004) Makeig et al. (2002).
Individual variability can be theoretically productive because it creates tests of how prior state constrains responsiveness. The relevant question is often whether a state measured before stimulation predicts a subsequent stimulus-specific neural or behavioral outcome. The temporal ordering alone does not prove an effect of state on behavior: common causes and selection processes may drive both. Randomized changes in task demand, perturbation, careful controls, and within-person replication can distinguish predictions more effectively than post hoc correlations. Multi-session data are especially valuable when the proposed feature is intended to characterize an individual Gross et al. (2013) Jas et al. (2018).
Statistical Inference Across Time, Channels, and Sources
Modern EEG/MEG data contain many correlated observations, including samples across time, frequency, sensors, anatomical estimates, and experimental conditions. Testing each point without correcting for multiplicity produces many opportunities for false discoveries. Maris and Oostenveld introduced nonparametric cluster-based procedures that incorporate the adjacency structure of electrophysiological observations and control relevant familywise inference under specified assumptions Maris & Oostenveld (2007). The cluster statistic evaluates a property of a collection of samples, rather than supplying a precise estimate of when and where a physiological effect begins.
Sassenhagen and Draschkow specifically clarified that cluster-based permutation tests do not establish inferential significance for the onset latency or location of an effect Sassenhagen & Draschkow (2019). A significant temporal cluster can justify a claim that conditions differ somewhere within the tested domain, while a statement about the earliest onset requires another estimator and uncertainty model. Likewise, an extensive topographic cluster does not automatically demonstrate a single anatomical generator. The manuscript of an EEG/MEG experiment should explicitly separate exploratory maps, confirmatory testing, and later localization claims.
Group-level inference also depends on the true unit of replication. Thousands of time samples or hundreds of trials from one participant cannot substitute for independent participants when a population claim is intended. Unequal trial counts, missing channels, participant-level exclusions, temporal autocorrelation, and clustered behavior can make naïve analyses misleading. Preregistration and transparent reporting of all comparisons reduce analytical flexibility, while hierarchical models can connect trial variability with participant-level effects when their assumptions are appropriate. Reporting effect estimates and uncertainty alongside threshold decisions improves scientific interpretation Gross et al. (2013) Keil et al. (2014).
Reproducibility, Reporting, and Open-Science Data Standards
Reproducibility in electrophysiology requires reconstructing both the physical recording and the chain of analysis. General labels such as 'EEG recorded and filtered' conceal crucial choices about electrode positions, referencing, timing synchronization, artifact rejection, and condition selection. The committee recommendations by Keil and colleagues and the MEG reporting framework led by Gross specify practical information needed to interpret and reproduce these studies Keil et al. (2014) Gross et al. (2013). Their role is methodological: they cannot guarantee valid science, but they make assumptions and experimental decisions available for evaluation.
The EEG-BIDS proposal establishes a structured convention for organizing recordings and metadata, including participants, sessions, task descriptions, channels, electrodes, and events Pernet et al. (2019). Analysis environments such as FieldTrip and MNE provide reusable implementations that can be tested across datasets Oostenveld et al. (2011) Gramfort et al. (2014). A reproducible study records software versions, dependencies, parameters, scripts, and data exclusions, and identifies which outputs are directly derived from which source files. This matters especially for multimodal projects because co-registration and source-space processing can otherwise become opaque.
Independent replication is a separate scientific achievement. Reproducing a figure with the same dataset demonstrates computational reproducibility; finding an analogous effect with new participants, equipment, or laboratories addresses replicability and generalization. Both are needed when proposing stable neural markers of cognition or clinical status. Sampling and measurement variability can produce different reported temporal signatures even under similar tasks, so researchers should explain the scope of any replication claim. Open data and code make cross-laboratory comparisons feasible while demanding appropriate consent, privacy protection, and responsible stewardship of identifiable recordings Jas et al. (2018) Pernet et al. (2019).
Perception and Temporal Sampling
Perceptual research uses the temporal resolution of EEG and MEG to investigate how sensory responses unfold after a stimulus and how ongoing activity changes subsequent processing. Early components, induced oscillations, and ongoing phase structure can each provide complementary constraints. Perceptual-cycle proposals consider whether some sensory operations fluctuate with the phase of neural rhythms VanRullen (2016). These proposals are scientifically interesting because they connect physiological timescales with precise psychophysical tasks, but a periodic behavioral modulation requires statistical validation beyond observing an oscillatory scalp signal.
A neural frequency can appear behaviorally relevant because it indexes sampling, modulates sensory gain, tracks anticipation, or covaries with broader brain state. Discriminating these accounts requires stimuli distributed across controlled phases or times, careful treatment of phase-estimation error, and explicit tests of competing models. Electrical and magnetic fields aggregate spatial sources, so the observed phase may not correspond to the particular pathway implementing the behavioral judgment. Alpha-related experimental work also illustrates that anticipatory attentional allocation can change later sensory responses Brignani et al. (2009) Jensen & Mazaheri (2010).
The main explanatory achievement is temporal constraint. If a task manipulation reliably changes a source-resolved response during a defined interval and predicts a corresponding behavioral contrast, candidate accounts of perception can be narrowed. The strength of that narrowing depends on trial counts, stimulus control, model choice, and independent replication. A timing signature should be understood as one level in an explanatory chain from sensory input through neural population dynamics to choice or report, with each transition supported by explicit evidence Luck et al. (2000) VanRullen (2016).
Attention, Memory, and Cognitive Selection
Selective attention changes how the nervous system responds to competing inputs, and EEG/MEG can locate experimental differences in time relative to cue, target, response, and feedback. ERP research has shown that different attentional manipulations alter temporally distinguishable components, while anticipatory spectral changes can precede relevant inputs Luck et al. (2000) Brignani et al. (2009). This evidence helps classify the timing of selection but remains dependent on task definitions. A contrast between attended and unattended trials can reflect differences in expectations, eye position, motivation, stimulus statistics, or motor preparation unless these are controlled.
Memory research can use oscillatory measures to test hypotheses about encoding, maintenance, and retrieval. Associations involving theta, alpha, and other bands have motivated accounts of temporal organization, but the same measured change may occur with effort, vigilance, or behavioral preparation. The task must distinguish memory load, representational content, delay length, and response demands before a recorded difference is assigned to memory itself. Individual-frequency variation and source uncertainty further limit conclusions from a standard band average Klimesch (2012) Cohen (2017).
Temporally resolved experiments are especially useful when a model predicts ordered stages or recurrent exchanges among systems. Yet a serial succession of scalp peaks can result from changing mixtures of overlapping generators rather than discrete stages. Comparing trial-level temporal patterns, explicit response models, and behavior can reveal whether a proposed sequence is distinguishable from continuous or recurrent processing. Functional claims gain credibility through convergent manipulations and independent replication, as required by general EEG/MEG reporting principles Gross et al. (2013) Keil et al. (2014).
Sensorimotor Preparation and Action Timing
Movement experiments connect temporally resolved electromagnetic signals to precisely specified motor events. Response-locked potentials, preparatory activity, event-related beta suppression, and postmovement oscillatory changes can be compared with movement onset, muscular activity, and trial-specific choices. Work on event-related synchronization and desynchronization provides a physiological framework for quantifying task-dependent changes in sensorimotor rhythms Pfurtscheller & Lopes da Silva (1999). MEG source models can further estimate cortical generators in motor systems, although the ability to identify a sensorimotor region does not alone distinguish motor command, sensory prediction, or proprioceptive feedback.
Experimental designs should differentiate internally initiated movements, instructed responses, cancellation, and externally imposed motion. Electromyography, kinematic monitoring, and reliable response timestamps help clarify when muscular execution begins relative to apparent cortical activity. The difficulty of interpreting movement-related features is increased by motion artifacts and the fact that changes in attention, arousal, and action preparation may coincide. OPM-MEG recordings expand the types of feasible movements in controlled settings, with the acquisition benefits and remaining interference challenges described in contemporary instrument studies Boto et al. (2018) Holmes et al. (2025).
The scientific goal is to establish which temporal events predict or constrain action under manipulations that selectively change preparation, commitment, or feedback. A group-level beta effect may be highly reproducible while individual prediction remains uncertain, and a peak that precedes average reaction time may reflect shared task timing rather than a necessary initiating process. Causal claims require interventions and controls that isolate their influence on behavioral performance. EEG and MEG are thus particularly strong at defining candidate temporal structure and testing model-based predictions about action sequences Gross et al. (2013) Cohen (2017).
Language and Social Brain Dynamics
Spoken language unfolds in continuous time, requiring auditory analysis, lexical access, contextual integration, prediction, and articulation across overlapping timescales. EEG and MEG can measure systematic changes following words, phonological features, or sentence-level events. However, a response attributed to language comprehension must be separated from auditory energy, task demands, eye movements, and general surprise. The inference depends on carefully matched stimuli and contrasts between specified linguistic operations, supported by established standards for interpreting event-related responses Luck et al. (2000) Gross et al. (2013).
Naturalistic speech presents a methodological challenge because acoustic, linguistic, and social features covary over time. Statistical response models can identify components tracking speech envelopes or other continuous predictors, but shared temporal structure may create associations without establishing a causal linguistic computation. Permutation and cross-validation strategies should preserve relevant autocorrelation and divide data so that adjacent segments do not leak information between training and testing. The field must distinguish decoding the stimulus from explaining how a particular participant processes its meaning Maris & Oostenveld (2007) Cohen (2017).
Research on social exchange adds turn taking, mutual adaptation, gaze, and multimodal sensory feedback. Simultaneous recordings from multiple people can reveal coordination, but synchrony can arise simply because participants experience common events or produce aligned actions. Demonstrating interpersonal neural coupling as a distinct explanatory process requires task-specific controls, delays consistent with hypothesized interaction, and evidence that neural measurements contribute beyond shared behavior. The problem parallels network-connectivity inference within one brain and demonstrates the general necessity of distinguishing covariation, coordination, and functional interaction Varela et al. (2001) Haufe et al. (2013).
Developmental and Clinical Translation
EEG's relative portability and accessibility have made it an important tool in studying developmental differences and clinical neurophysiology, while MEG provides complementary source-sensitive measurements in suitable settings. Comparisons across age, population, and diagnosis are especially vulnerable to differences in head anatomy, movement, compliance, sleep, vigilance, and signal strength. When groups show different average oscillatory power or component latency, investigators must evaluate measurement equivalence before interpreting the contrast as a change in a distinct psychological mechanism Keil et al. (2014) Baillet (2017).
Clinical studies investigate whether temporally resolved features relate to symptoms, function, risk, or treatment-associated changes. A group difference may have scientific value while offering weak performance for diagnosis of an individual. Biomarker claims require prespecified outcomes, reliable repeated measurements, external validation, appropriate comparison groups, and an analysis of whether a proposed feature improves predictions beyond readily available clinical information. The reproducibility framework and good-practice guidance for EEG/MEG are necessary foundations for any translational program Jas et al. (2018) Gross et al. (2013).
New wearable systems create opportunities for research in populations previously difficult to scan in fixed MEG devices. A 2025 study of repeated OPM-MEG measurements explicitly investigated the stability of evoked, induced, and passive responses, underscoring the need to estimate reliability separately for distinct features and recording conditions Rhodes et al. (2025). Its scientific significance lies in an empirically testable reliability question, rather than a general assertion that each measurement is already a validated clinical marker. Clinical translation should proceed through independent cohort replication, measurement standardization, and appropriate professional oversight.
Wearable Magnetometry and Ecological Measurement
Optically pumped magnetometers have made it possible to build MEG arrays that can be worn and adapted to different head geometries. In a foundational demonstration, Boto and colleagues used a wearable system to advance the recording of brain responses during movement Boto et al. (2018). Later engineering work investigated optical sensor operation, array design, and the tradeoffs among mobility, sensor proximity, field interference, and spatial localization Tierney et al. (2019). Such developments change which psychological tasks can be studied, but a wearable array remains dependent on accurate sensor positions and carefully characterized magnetic environments.
Experimental reports published in 2024–2025 address specific validation requirements. Seedat and colleagues examined concurrent whole-head EEG and OPM-MEG in motor and eyes-open/eyes-closed tasks, providing evidence about signal quality within their experimental arrangements Seedat et al. (2024). Holmes and colleagues compared recordings in different shielding conditions, using signal-space separation methods to reduce interference in a lightly shielded environment Holmes et al. (2025). Hill and colleagues studied estimation of sensor geometry and gain, which are essential for reliable source imaging Hill et al. (2025). These studies expand the engineering evidence base without warranting generalization to every population, location, or naturally moving task.
Naturalistic designs introduce new explanatory opportunities and new confounds. Movement changes the relation between sources and sensors, while magnetic interference, physiological artifact, task variability, and behavioral context become less controlled. The appropriate standard is measurement validation matched to the intended experiment: phantom tests, repeated measures, source localization benchmarks, simultaneous behavioral sensing, and sensitivity analysis for data-processing choices. Advances in instrumentation only become advances in psychological knowledge when new experiments discriminate among substantive hypotheses about neural and behavioral timing Rhodes et al. (2025) Gross et al. (2013).
Critical Scientific Discussion: Causal and Construct Validity
EEG and MEG provide exceptionally informative time-resolved access to changes in neural population activity, but the resulting explanation is mediated by a succession of models. Physical sensors record fields; acquisition and preprocessing select observable features; source models estimate candidate generators; experimental contrasts associate features with tasks; and theoretical analysis attributes psychological functions. Errors or untested assumptions at any transition can change the final interpretation. The strength of a result therefore depends on specifying which transition is empirically supported, and which remains a hypothesis Nunez & Srinivasan (2006) Baillet (2017).
Three recurrent inferential errors deserve special attention. The first is treating named frequency bands or waveform peaks as unique psychological processes. The second is treating a reconstructed source or connectivity edge as an observed anatomical interaction. The third is treating statistical significance of a broad time-frequency cluster as proof of an exact event onset or brain location. The measurement and methodological literature provides direct grounds for avoiding these errors, including source-mixing simulations, models of periodic and aperiodic power, and explicit limits of permutation inference Donoghue et al. (2020) Haufe et al. (2013) Sassenhagen & Draschkow (2019).
Better explanation comes from experimental convergence: accurately timed behavioral tasks, preregistered contrasts, signal simulations, valid source geometry, independent replication, and—where appropriate—selective perturbation. Predictive models can be useful, but their out-of-sample accuracy must be assessed separately from causal interpretation. A signal that predicts reaction time across participants can be scientifically important while remaining compatible with several neural mechanisms. The conceptual achievement of temporal neurophysiology lies in making those alternative mechanisms testable and narrowing their plausible timing, without assuming that temporal proximity identifies a complete cause Fries (2005) Gross et al. (2013).
Open Research Problems and Future Directions
A major open problem is disentangling rhythms from transient structured events and changes in aperiodic activity at the level of single trials and sources. Multiple analysis pipelines can generate similar grand-average patterns while implying different neural dynamics. Future experiments should compare model predictions for burst timing, spectral peaks, phase stability, and the spatial distribution of underlying generators, rather than relying on one descriptive summary. Quantifying uncertainty in these estimates and preserving raw data will strengthen their interpretability across laboratories Donoghue et al. (2020) Pernet et al. (2019).
A second problem concerns the interpretation of large-scale neural interactions. Source leakage, common inputs, network feedback, and physiological nonstationarity make it difficult to identify directed communication from passive data. Stronger designs will combine cross-modal convergence with temporal interventions and realistic simulations of each sensing modality. Population-level generalization requires participants across demographic, developmental, and clinical contexts while testing whether measurement models remain comparable across groups. Recent advances in wearable MEG also raise the practical challenge of transporting validated methods into more mobile conditions without sacrificing known error bounds Holmes et al. (2025) Hill et al. (2025).
A third problem is connecting temporal neural descriptions with precise psychological theory. Competing accounts of attention, memory, decision, and perception often predict broad activity differences yet fail to specify distinct temporal signatures. Research gains explanatory power when theories state what will change, when it will change, in which generators, and how that change will appear in behavior under experimental manipulation. This creates a common methodological foundation for neuroscience in the Artificial Era: claims about biological and engineered temporal processing should use operational definitions, measurement models, and explicit causal tests, while remaining clear about the different physical substrates and evidence available Cohen (2017) VanRullen (2016).
Conclusion
EEG, MEG, and the study of neural temporal dynamics form a coherent scientific field concerned with the measurable timing of brain population activity and its relationship to cognition and behavior. The historical development from early human EEG to multichannel MEG and modern wearable systems has progressively expanded the range of states and activities accessible to noninvasive measurement. Contemporary analysis makes event-related responses, oscillations, transient bursts, phase relations, and reconstructed sources available as distinct observational objects, each governed by its own statistical and biophysical assumptions Berger (1929) Hämäläinen et al. (1993) Baillet (2017).
The evidence supports a principled distinction between temporal neural signatures and the psychological mechanisms proposed to explain them. ERP components constrain response timing; power spectra characterize rhythmic and aperiodic structure; connectivity statistics quantify dependence under conditions of source mixing; and inverse solutions supply conditional anatomical estimates. Their explanatory value depends on valid experimental contrasts, carefully described signal acquisition, reproducible preprocessing, appropriate statistics, and behavioral definitions. The strongest findings survive alternative analyses and independent replication while making clear the limitations of inferences about exact onset, location, or causal direction Keil et al. (2014) Maris & Oostenveld (2007) Sassenhagen & Draschkow (2019).
The resulting synthesis establishes an integrated research program centered on biophysical observability, temporal characterization, source-informed inference, and psychological construct validation. Open questions about neural bursts, network communication, clinical measurement reliability, and naturalistic acquisition remain tractable through increasingly transparent and convergent methods. A rigorous future for EEG and MEG advances alongside, and helps discipline, larger theoretical claims about cognition: precise temporal measurement becomes explanatory knowledge when the hypothesized relationship among neural mechanism, task conditions, and behavior withstands discriminating empirical tests Gross et al. (2013) Rhodes et al. (2025).
References
- Acunzo, D. J., MacKenzie, G., & van Rossum, M. C. W. (2012). Systematic biases in early ERP and ERF components as a result of high-pass filtering. Journal of Neuroscience Methods, 209(1), 212–218. https://doi.org/10.1016/j.jneumeth.2012.06.011
- Arieli, A., Sterkin, A., Grinvald, A., & Aertsen, A. (1996). Dynamics of ongoing activity: Explanation of the large variability in evoked cortical responses. Science, 273(5283), 1868–1871. https://doi.org/10.1126/science.273.5283.1868
- Baillet, S. (2017). Magnetoencephalography for brain electrophysiology and imaging. Nature Neuroscience, 20(3), 327–339. https://doi.org/10.1038/nn.4504
- Berger, H. (1929). Über das Elektrenkephalogramm des Menschen. Archiv für Psychiatrie und Nervenkrankheiten, 87, 527–570. https://doi.org/10.1007/BF01797193
- Boto, E., Holmes, N., Leggett, J., Roberts, G., Shah, V., Meyer, S. S., Muñoz, L. D., Mullinger, K. J., Tierney, T. M., Bestmann, S., Barnes, G. R., Bowtell, R., & Brookes, M. J. (2018). Moving magnetoencephalography towards real-world applications with a wearable system. Nature, 555, 657–661. https://doi.org/10.1038/nature26147
- Brignani, D., Lepsien, J., Rushworth, M. F. S., & Nobre, A. C. (2009). The timing of neural activity during shifts of spatial attention. Journal of Cognitive Neuroscience, 21(12), 2369–2383. https://doi.org/10.1162/jocn.2008.21176
- Cohen, M. X. (2017). Where does EEG come from and what does it mean? Trends in Neurosciences, 40(4), 208–218. https://doi.org/10.1016/j.tins.2017.02.004
- Dale, A. M., Liu, A. K., Fischl, B. R., Buckner, R. L., Belliveau, J. W., Lewine, J. D., & Halgren, E. (2000). Dynamic statistical parametric mapping: Combining fMRI and MEG for high-resolution imaging of cortical activity. Neuron, 26(1), 55–67. https://doi.org/10.1016/S0896-6273(00)81138-1
- Delorme, A., & Makeig, S. (2004). EEGLAB: An open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. Journal of Neuroscience Methods, 134(1), 9–21. https://doi.org/10.1016/j.jneumeth.2003.10.009
- Donoghue, T., Haller, M., Peterson, E. J., Varma, P., Sebastian, P., Gao, R., Noto, T., Lara, A. H., Wallis, J. D., Knight, R. T., Shestyuk, A., & Voytek, B. (2020). Parameterizing neural power spectra into periodic and aperiodic components. Nature Neuroscience, 23(12), 1655–1665. https://doi.org/10.1038/s41593-020-00744-x
- Fries, P. (2005). A mechanism for cognitive dynamics: Neuronal communication through neuronal coherence. Trends in Cognitive Sciences, 9(10), 474–480. https://doi.org/10.1016/j.tics.2005.08.011
- Gramfort, A., Luessi, M., Larson, E., Engemann, D. A., Strohmeier, D., Brodbeck, C., Goj, R., Jas, M., Brooks, T., Parkkonen, L., & Hämäläinen, M. S. (2014). MNE software for processing MEG and EEG data. NeuroImage, 86, 446–460. https://doi.org/10.1016/j.neuroimage.2013.10.027
- Gross, J., Baillet, S., Barnes, G. R., Henson, R. N., Hillebrand, A., Jensen, O., Jerbi, K., Litvak, V., Maess, B., Oostenveld, R., Parkkonen, L., Taylor, J. R., van Wassenhove, V., Wibral, M., & Schoffelen, J.-M. (2013). Good practice for conducting and reporting MEG research. NeuroImage, 65, 349–363. https://doi.org/10.1016/j.neuroimage.2012.10.001
- Hämäläinen, M., Hari, R., Ilmoniemi, R. J., Knuutila, J., & Lounasmaa, O. V. (1993). Magnetoencephalography—theory, instrumentation, and applications to noninvasive studies of the working human brain. Reviews of Modern Physics, 65(2), 413–497. https://doi.org/10.1103/RevModPhys.65.413
- Hari, R., Parkkonen, L., & Nangini, C. (2010). The brain in time: Insights from neuromagnetic recordings. Annals of the New York Academy of Sciences, 1191, 89–109. https://doi.org/10.1111/j.1749-6632.2010.05438.x
- Haufe, S., Nikulin, V. V., Müller, K.-R., & Nolte, G. (2013). A critical assessment of connectivity measures for EEG data: A simulation study. NeuroImage, 64, 120–133. https://doi.org/10.1016/j.neuroimage.2012.09.036
- Hill, R. M., Reina Rivero, G., Tyler, A. J., Schofield, H., Doyle, C., Osborne, J., Bobela, D., Rier, L., Gibson, J., Tanner, Z., Boto, E., Bowtell, R., Brookes, M. J., & Shah, V. (2025). Determining sensor geometry and gain in a wearable MEG system. Imaging Neuroscience, 3, imag_a_00535. https://doi.org/10.1162/imag_a_00535
- Holmes, N., Leggett, J., Hill, R. M., Rier, L., Boto, E., Schofield, H., Hayward, T., Dawson, E., Woolger, D., Shah, V., Taulu, S., Brookes, M. J., & Bowtell, R. (2025). Wearable magnetoencephalography in a lightly shielded environment. IEEE Transactions on Biomedical Engineering, 72(2), 609–618. https://doi.org/10.1109/TBME.2024.3465654
- Jas, M., Larson, E., Engemann, D. A., Leppäkangas, J., Taulu, S., Hämäläinen, M., & Gramfort, A. (2018). A reproducible MEG/EEG group study with the MNE software: Recommendations, quality assessments, and good practices. Frontiers in Neuroscience, 12, Article 530. https://doi.org/10.3389/fnins.2018.00530
- Jensen, O., & Mazaheri, A. (2010). Shaping functional architecture by oscillatory alpha activity: Gating by inhibition. Frontiers in Human Neuroscience, 4, Article 186. https://doi.org/10.3389/fnhum.2010.00186
- Keil, A., Debener, S., Gratton, G., Junghöfer, M., Kappenman, E. S., Luck, S. J., Luu, P., Miller, G. A., & Yee, C. M. (2014). Committee report: Publication guidelines and recommendations for studies using electroencephalography and magnetoencephalography. Psychophysiology, 51(1), 1–21. https://doi.org/10.1111/psyp.12147
- Klimesch, W. (2012). Alpha-band oscillations, attention, and controlled access to stored information. Trends in Cognitive Sciences, 16(12), 606–617. https://doi.org/10.1016/j.tics.2012.10.007
- Luck, S. J., Woodman, G. F., & Vogel, E. K. (2000). Event-related potential studies of attention. Trends in Cognitive Sciences, 4(11), 432–440. https://doi.org/10.1016/S1364-6613(00)01545-X
- Makeig, S., Westerfield, M., Jung, T.-P., Enghoff, S., Townsend, J., Courchesne, E., & Sejnowski, T. J. (2002). Dynamic brain sources of visual evoked responses. Science, 295(5555), 690–694. https://doi.org/10.1126/science.1066168
- Maris, E., & Oostenveld, R. (2007). Nonparametric statistical testing of EEG- and MEG-data. Journal of Neuroscience Methods, 164(1), 177–190. https://doi.org/10.1016/j.jneumeth.2007.03.024
- Nolte, G., Bai, O., Wheaton, L., Mari, Z., Vorbach, S., & Hallett, M. (2004). Identifying true brain interaction from EEG data using the imaginary part of coherency. Clinical Neurophysiology, 115(10), 2292–2307. https://doi.org/10.1016/j.clinph.2004.04.029
- Nunez, P. L., & Srinivasan, R. (2006). Electric fields of the brain: The neurophysics of EEG (2nd ed.). Oxford University Press. https://doi.org/10.1093/acprof:oso/9780195050387.001.0001
- Oostenveld, R., Fries, P., Maris, E., & Schoffelen, J.-M. (2011). FieldTrip: Open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data. Computational Intelligence and Neuroscience, 2011, Article 156869. https://doi.org/10.1155/2011/156869
- Pernet, C. R., Appelhoff, S., Gorgolewski, K. J., Flandin, G., Phillips, C., Delorme, A., & Oostenveld, R. (2019). EEG-BIDS, an extension to the brain imaging data structure for electroencephalography. Scientific Data, 6, Article 103. https://doi.org/10.1038/s41597-019-0104-8
- Pfurtscheller, G., & Lopes da Silva, F. H. (1999). Event-related EEG/MEG synchronization and desynchronization: Basic principles. Clinical Neurophysiology, 110(11), 1842–1857. https://doi.org/10.1016/S1388-2457(99)00141-8
- Rhodes, N., Bandhan, J., Vandewouw, M. M., Coleman, S. C., & Taylor, M. J. (2025). Assessing the stability of evoked, induced, and passive MEG responses for repeat testing with optically pumped magnetometers. Scientific Reports, 15, Article 26722. https://doi.org/10.1038/s41598-025-12131-2
- Sassenhagen, J., & Draschkow, D. (2019). Cluster-based permutation tests of MEG/EEG data do not establish significance of effect latency or location. Psychophysiology, 56(6), e13335. https://doi.org/10.1111/psyp.13335
- Seedat, Z. A., St. Pier, K., Holmes, N., Rea, M., Al-Hilaly, L., Tierney, T. M., Embury, C. M., Pardington, R., Mullinger, K. J., Cross, J. H., Boto, E., & Brookes, M. J. (2024). Simultaneous whole-head electrophysiological recordings using EEG and OPM-MEG. Imaging Neuroscience, 2, 1–15. https://doi.org/10.1162/imag_a_00179
- Tallon-Baudry, C., Bertrand, O., Delpuech, C., & Pernier, J. (1996). Stimulus specificity of phase-locked and non-phase-locked 40 Hz visual responses in human. Journal of Neuroscience, 16(13), 4240–4249. https://doi.org/10.1523/JNEUROSCI.16-13-04240.1996
- Taulu, S., & Simola, J. (2006). Spatiotemporal signal space separation method for rejecting nearby interference in MEG measurements. Physics in Medicine & Biology, 51(7), 1759–1768. https://doi.org/10.1088/0031-9155/51/7/008
- Tierney, T. M., Holmes, N., Mellor, S., López, J. D., Roberts, G., Hill, R. M., Boto, E., Leggett, J., Shah, V., Brookes, M. J., Bowtell, R., & Barnes, G. R. (2019). Optically pumped magnetometers: From quantum origins to multi-channel magnetoencephalography. NeuroImage, 199, 598–608. https://doi.org/10.1016/j.neuroimage.2019.05.063
- Van Veen, B. D., van Drongelen, W., Yuchtman, M., & Suzuki, A. (1997). Localization of brain electrical activity via linearly constrained minimum variance spatial filtering. IEEE Transactions on Biomedical Engineering, 44(9), 867–880. https://doi.org/10.1109/10.623056
- VanRullen, R. (2016). Perceptual cycles. Trends in Cognitive Sciences, 20(10), 723–735. https://doi.org/10.1016/j.tics.2016.07.006
- Varela, F., Lachaux, J.-P., Rodriguez, E., & Martinerie, J. (2001). The brainweb: Phase synchronization and large-scale integration. Nature Reviews Neuroscience, 2(4), 229–239. https://doi.org/10.1038/35067550
- Vinck, M., Oostenveld, R., van Wingerden, M., Battaglia, F., & Pennartz, C. M. A. (2011). An improved index of phase-synchronization for electrophysiological data in the presence of volume-conduction, noise and sample-size bias. NeuroImage, 55(4), 1548–1565. https://doi.org/10.1016/j.neuroimage.2011.01.055
- Widmann, A., Schröger, E., & Maess, B. (2015). Digital filter design for electrophysiological data—a practical approach. Journal of Neuroscience Methods, 250, 34–46. https://doi.org/10.1016/j.jneumeth.2014.08.002