What EEG measures

Why this matters

Everything you will measure in this curriculum is a voltage difference between two points on the skin, produced by currents that flow inside a head whose skull conducts electricity poorly. That single fact explains why EEG is fast but blurry, why a “source” reported from a scalp map deserves suspicion, why neighbouring electrodes always look alike, and why a microvolt signal has to be dug out of millivolt-scale offsets. If you know where the signal comes from, most of the later warnings in this curriculum become predictions rather than rules.

Concepts

The generator: postsynaptic currents in aligned pyramidal cells

Scalp EEG is not a recording of action potentials. Spikes are brief (about a millisecond), and the spikes of different neurons are not aligned in time, so their fields cancel at a distance. What survives at the scalp is the sum of slower postsynaptic potentials: when a synapse on a pyramidal cell’s apical dendrite is active, current flows into the dendrite at one place (a current sink) and out of the cell elsewhere along its length (a source). Seen from a distance, a sink and a source separated along the dendrite behave like a small current dipole.

A single cell’s dipole is far too weak to detect at the scalp. Pyramidal cells in cortex, however, are arranged in a palisade: their apical dendrites run in parallel, perpendicular to the cortical surface. When many thousands of them receive similar input at the same time, their dipoles point the same way and add. The textbook estimate is that a patch of cortex on the order of several square centimetres, active in near-synchrony, is needed to produce a potential measurable at the scalp (TODO(confirm) the figure the author prefers; Nunez and Srinivasan give the standard treatment). Two things therefore decide what reaches the scalp: synchrony (are the sinks and sources aligned in time?) and geometry (are the dipoles aligned in space?).

Open and closed fields

Geometry can defeat even perfect synchrony. A population whose dipoles all point in the same direction produces an open field: its potentials add and can be measured far away. A population whose dipoles point in all directions, or in a radially symmetric arrangement (as in some nuclei), produces a closed field: the potentials cancel outside the structure, however active it is. This is why cortex dominates EEG and why deep, non-laminar structures contribute little that can be attributed to them from the scalp.

The equivalent current dipole

For modelling, a synchronously active cortical patch is replaced by one equivalent current dipole with a position, an orientation and a strength (the dipole moment, in nanoampere-metres). The orientation is set by the cortical surface, because the dendrites are perpendicular to it:

• A radial dipole (on a gyral crown, pointing at the skull) produces a scalp map with one extremum directly above it, of one sign, fading smoothly with distance.
• A tangential dipole (on a sulcal wall, parallel to the skull) produces a pair of extrema of opposite sign on either side of the source, with a potential near zero directly above it. A tangential source can therefore look as if it came from two places, neither of which is where it is.
• A deeper dipole produces a broader and weaker map. Its map can resemble that of a weaker, shallower source, which is one root of the overclaiming that the pitfall below describes.

Because the potentials of several dipoles add linearly, the scalp map of two simultaneous sources is the sum of their two maps, and a map alone cannot tell you how many sources made it. The widget lets you build exactly this intuition: one dipole, rotate it, push it deeper, then add a second.

Volume conduction and the skull as a spatial low-pass filter

Currents do not travel from the cortex to the electrode along a wire; they spread through the brain, the cerebrospinal fluid, the skull and the scalp, a process called volume conduction. Each tissue conducts differently, and the skull conducts far worse than the tissues on either side of it (the skull-to-brain conductivity ratio is itself a debated quantity; TODO(confirm) the range the author wants stated). Two consequences follow:

1. Attenuation. Potentials at the scalp are much smaller than those on the cortical surface.
2. Spatial blur. The poorly conducting skull, sandwiched between the well-conducting CSF and scalp, spreads current sideways before it reaches the electrodes. Fine spatial detail in the cortical potential is smoothed away: the skull acts as a spatial low-pass filter. Every electrode sees a weighted sum of many sources, and neighbouring electrodes are strongly correlated whether or not the underlying cortex is.

Volume conduction is the central problem of Level 5 (connectivity and source analysis): sensors that look “connected” at zero lag are usually just seeing the same source. For now the lesson is simpler: a scalp map is a blurred, attenuated, superposed picture of cortical currents, and reading it backwards to the cortex is an inverse problem with no unique solution.

The scale of the signal

Scalp EEG lives in microvolts. Ongoing rhythms such as posterior alpha are typically tens of microvolts peak to peak; event-related potentials are a few microvolts and need averaging to see (TODO(confirm) the ranges the author wants quoted). Intracranial recordings from the cortical surface or from depth electrodes are larger by orders of magnitude, because there is no skull between the electrode and the source: the catalog notes that the scalp and intracranial sets of ds-bonn differ by orders of magnitude in amplitude. Meanwhile the electrode–skin interface carries slowly varying offsets on the scale of millivolts. The amplifier’s job (L0.3) is to reject what is common to both inputs and to resolve microvolt differences riding on millivolt offsets.

Temporal resolution and the resolution ladder

What EEG loses in space it keeps in time. The postsynaptic currents it records change on the scale of milliseconds, and the recording follows them directly, with no slow intermediary such as blood flow. Placing EEG among its neighbours:

The rungs of this ladder are anchored by real recordings of the same kind of resting rhythm at three scales: scalp EEG (ds-eegbci), human intracranial EEG (ds-hup or ds-respect) and a rodent high-density array (ds-mouse). The amplitude scale on each panel is part of the lesson.

The data behind this lesson

• ds-eegbci supplies the scalp panel (64-channel cap, 160 Hz, no hardware filters; ODC-By 1.0 on PhysioNet, CC0 on the OpenNeuro mirror).
• ds-hup and ds-respect are human intracranial datasets (CC0). They appear only as precomputed figures; no intracranial snippet ships on the site, and their patient-specific montages have no fixed channel count.
• ds-mouse (CC BY 4.0) supplies the rodent high-density-array panel, again as a precomputed figure; note its recordings already carry a 60 Hz notch and a 0.3–100 Hz analog band-pass.
• ds-bonn would give the scalp-versus-intracranial contrast inside one dataset (sets A/B scalp, C/D/E intracranial), but its terms are informal and no Bonn-derived asset ships unless §13 item 19 permits it (TODO(confirm)).

Explore

Work through it in this order: one radial dipole under Cz, then rotate it by ninety degrees and watch the single maximum become a pair; push it deeper and watch the map broaden and weaken; then add a second dipole and look for the map that neither would make alone. The head model here is a homogeneous single-shell sphere with no separate skull layer, so real scalp maps are blurrier than these; realistic head models arrive with the forward mode in Phase 3. The “match the maps” task at the end of the widget is this lesson’s exercise.

Practice

This lesson has no notebook. Its practice is the widget’s match-the-maps task below, and the first hands-on notebook of the level, nb-0-2-montages, follows in L0.2 once you know the names of the electrodes you will be placing.

Exercises