EEG Signal Generation

Geometry is destiny: the pyramidal palisade

The reason the human cortex produces a recordable EEG at all is architectural. Cortical pyramidal neurons are arranged in palisades - long apical dendrites running in parallel, perpendicular to the cortical surface, like a field of vertical antennae. When a population of these cells receives synchronous synaptic input, their individual transmembrane currents do not point in random directions; they all align along the same radial axis. Alignment is the precondition for summation. A structure in which the generators are geometrically parallel is called an open field, and only open fields project a current to a distance. Brain regions whose neurons are radially symmetric or randomly oriented form closed fields whose currents cancel locally and contribute little or nothing to the scalp signal. This is why the cerebral cortex, with its exquisitely layered, oriented pyramidal architecture, is the dominant generator of the EEG, while many deep nuclei with stellate or spherical cell geometries are nearly invisible.

Consider a single excitatory synapse onto the superficial apical dendrite. Cations flow into the cell at that point - a current sink - and an equal current must flow out across the membrane elsewhere, near the soma and basal dendrites - a current source. This conservation of current is not optional; it is Kirchhoff's law applied to a cell, and it is the reason a dipole always has two poles. The spatial separation of sink and source along the dendritic axis constitutes an electrical dipole: a tiny vector pointing, by convention, from the negative to the positive pole. Because neighboring pyramidal cells are parallel and are driven together, their dipoles stack into a coherent sheet, an extended dipole layer that behaves, from the distance of the scalp, like a single equivalent dipole. The strength of a single neuron's contribution is captured by its current dipole moment, on the order of a fraction of a picoampere-meter - roughly 0.1 to 1 pA-m (that is, hundreds of femto-ampere-meters) for a single pyramidal cell's postsynaptic event - vanishingly small individually, which is exactly why summation across tens of thousands of cells is not a refinement but a requirement.

Polarity: reading depth and sign from the surface

The polarity recorded at the scalp encodes both the depth of the active synapse and its sign. Four canonical combinations recur constantly in interpretation. Superficial excitation draws current inward near the surface, leaving a current sink at the top of the column and yielding a surface-negative field. Deep excitation places the sink near the soma and the source superficially, yielding a surface-positive field. Inhibition inverts the logic: superficial inhibition (outward current near the surface) produces surface positivity, while deep inhibition produces surface negativity. Thus a given scalp polarity is fundamentally ambiguous about mechanism - a surface-negative wave could be superficial excitation or deep inhibition - and resolving that ambiguity requires additional physiological context, such as the known laminar targets of the input system in question.

Summation, synchrony, and the area threshold

A single pyramidal dipole produces a field measured in nanovolts at the cortical surface and is utterly undetectable from the scalp. Recordability depends on summation, which has two requirements. First, temporal summation: the slow time course of postsynaptic currents (tens to hundreds of milliseconds) lets many events overlap in time. Second, spatial summation: parallel geometry lets many cells add coherently rather than cancel. The decisive variable is synchrony - the degree to which the population is driven in phase. A classic empirical benchmark, derived from intracranial-to-scalp correlation studies, holds that roughly 6 square centimeters of synchronously active cortex is a lower bound below which an interictal discharge essentially never reaches the scalp, while reliable scalp visibility generally requires more than 10 square centimeters - in one widely cited series about 90 percent of cortical spikes arising from a source area greater than 10 square centimeters produced a scalp-detectable spike, versus only about 10 percent of those from a smaller area. Many discrete events require even larger cooperating areas (10 to 20 square centimeters or more) to be recognized at the scalp. This is why a great deal of focal epileptiform activity that is obvious on intracranial electrodes never appears at the scalp at all - a sobering limit that shapes the entire practice of presurgical evaluation.

This synchrony-dominance has a profound and counterintuitive clinical consequence. Scalp amplitude is governed more by how coherently cortex fires than by how much total neural activity is present. A drowsy or encephalopathic brain whose neurons fall into slow, highly synchronous delta can generate a larger scalp signal than a healthy, alert, vigorously processing cortex whose neurons are richly desynchronized. High EEG amplitude is therefore not a marker of a healthy or active brain - it is a marker of synchrony, and in many contexts of pathological synchrony. Conversely, the low-amplitude, fast, desynchronized activation pattern of the alert state reflects efficient, decorrelated cortical processing. The single most important reframing for a trainee is this: more synchrony usually means less useful computation, and the dramatic, high-voltage records are often the sickest.

Synchrony itself is not a primitive; it is imposed by network mechanisms. Thalamocortical loops, with the reciprocal interplay between thalamic relay cells, the inhibitory reticular nucleus, and cortex, act as pacemakers that entrain large pyramidal populations - the source of spindles and of the spike-wave rhythm. Within cortex, fast-spiking inhibitory interneurons coupled by gap junctions and feedback inhibition rhythmically silence and release pyramidal cells, generating the gamma and other fast rhythms. The point for signal generation is that synchrony is manufactured by inhibition as much as by excitation: it is the rhythmic gating of pyramidal firing by interneuron networks that paces the macroscopic field. When you read a rhythm, you are reading the clock, not just the activity it gates.

Volume conduction: the head as a smearing filter

Between the cortical dipole layer and the recording electrode lie cerebrospinal fluid, dura, skull, and scalp - tissues of sharply differing conductivity. The current spreads passively through this inhomogeneous, layered conductor, a process called volume conduction. Because the skull is a poor conductor sandwiched between two good ones (CSF and scalp), it acts as a spatial low-pass filter: it blurs the field over several centimeters and attenuates it by roughly an order of magnitude relative to the cortical surface. The conductivity contrast across the skull is large - the skull is far more resistive than the soft tissue on either side - and this single anatomical fact, more than any other, is responsible for the limited spatial resolution of scalp EEG. The practical upshot is that each scalp electrode reports a spatially smeared average over a broad cortical region rather than a focal point.

A second, often-underappreciated consequence of volume conduction is that it is instantaneous and linear. Current from a generator reaches all electrodes essentially simultaneously, with no propagation delay across the head, because the relevant electromagnetic propagation is effectively immediate at these scales (the quasi-static approximation of Maxwell's equations holds for EEG). This is why two electrodes can show perfectly zero-phase-lag correlated activity simply because they both see the same source through volume conduction - a critical confound for any analysis of connectivity or synchrony between scalp channels. Apparent coupling between two regions may be nothing more than two windows onto a single generator, and distinguishing true neural interaction from shared volume-conducted source is one of the central methodological problems of quantitative EEG, addressed by measures designed to be insensitive to zero-lag coupling.

Gyral geometry and the tangential-source blind spot

The radial-antenna picture is an idealization that the folded cortex complicates in a way every expert must hold in mind. Pyramidal cells are oriented perpendicular to the local cortical surface, but the cortex is convoluted into gyri and sulci, so the dipole orientation depends on whether the active patch sits on a gyral crown or a sulcal wall. A generator on a gyral crown produces a radial dipole pointing straight out at the nearest electrode and yields a clean, well-localized scalp maximum. A generator on a sulcal wall produces a tangential dipole pointing sideways, whose scalp expression is split into a positive and a negative pole straddling the sulcus, with little signal directly above it - and a generator deep in a sulcal fundus may largely cancel against the opposing wall. The upshot is a systematic blind spot: a substantial fraction of cortex, buried in sulci and mesial surfaces, is poorly or misleadingly represented at the scalp.

This geometry explains many otherwise puzzling clinical observations. A mesial temporal or interhemispheric (mesial frontal) generator may produce a scalp field that is paradoxically maximal at the vertex or on the wrong side, because its tangential orientation projects the field away from the source. Frontopolar and basal temporal generators are notoriously under-detected. When the scalp localization and the clinical or imaging localization disagree, the resolution is frequently gyral geometry rather than error - a tangential source faithfully obeying physics in a way that defeats naive surface reading. This is also why magnetoencephalography (MEG), which is selectively sensitive to tangential sources and far less distorted by the skull, is complementary to EEG rather than redundant: the two modalities see partly different populations precisely because of orientation.

The inverse problem and forward modeling in 2026

Putting all of this together yields the central mathematical fact of source analysis: the inverse problem - inferring the configuration of cortical generators from the scalp map - has no unique solution. This is not a practical limitation that better hardware will overcome; it is a theorem. Helmholtz showed in the nineteenth century that infinitely many internal current distributions can produce identical surface fields, because a given external field can always be reproduced by different combinations of sources. The corollary is that source localization is impossible without constraints - prior assumptions that select a plausible solution from the infinite set. Common constraints include limiting solutions to a small number of equivalent current dipoles, requiring the smoothest distributed solution, restricting sources to the gray-matter surface derived from the patient's own MRI, or imposing sparsity. Each algorithm encodes a different prior, and different priors yield different answers from the same data.

Making any inverse solution meaningful requires first solving the easier forward problem: given an assumed source, predict the scalp field. This demands a head model that captures the conductive geometry. The state of the art has progressed from simple concentric-sphere approximations to realistic, MRI-derived boundary-element and finite-element models that represent the actual shape and conductivity of scalp, skull, CSF, and brain - and increasingly incorporate skull anisotropy and individualized conductivity estimation, since uncertainty in skull conductivity remains one of the dominant sources of localization error. The accuracy of a forward model sets a hard ceiling on the accuracy of any inverse solution built on it; a beautifully sophisticated source-localization algorithm fed an inaccurate head model produces confident, precise, and wrong answers. High-density arrays of 64 to 256 electrodes, co-registered to the individual's MRI, define current best practice for research-grade and presurgical source localization.

These principles are not abstract: every real recording is a volume-conducted, geometrically filtered shadow of cortical current. The posterior dominant rhythm provides the cleanest demonstration, because its generators in occipito-parietal cortex are favorably oriented and richly synchronous, producing a large, reactive field. Watch how the alpha rhythm behaves below, and read it as the visible output of the entire generative chain - synchronous synaptic currents on aligned pyramidal palisades, summed into a dipole layer, smeared by the skull, and modulated within a second by a change in cortical state.