Electroencephalography

Berger’s recordings were an amplifier problem before they were a neuroscience result. Scalp potentials are tens of microvolts, sitting on top of electrode drift, mains interference, and muscle activity orders of magnitude larger. That he saw the alpha rhythm at all in the 1920s is largely a story about galvanometer sensitivity and careful electrode contact.

Galvanometers first, amplifiers later

The currents were nearly fifty years old by then. In 1875 Richard Caton reported to the British Medical Association that his galvanometer had found currents in every rabbit and monkey brain he had examined — feeble ones, of varying direction, between points on the surface — and that grey matter at work usually swung negative: in the rabbit, the area Ferrier had tied to the eyelids responded when light fell on the opposite retina. Others found the currents independently, among them Adolf Beck in Kraków, who reported in 1890 that a flash of light or a handclap stopped their spontaneous fluctuations. All of that was exposed cortex in animals. Berger wanted the human brain through its coverings.

His human recordings began in 1924 with Edelmann string galvanometers, mostly on patients with openings in the skull, and from 1926 he used a Siemens double-coil galvanometer that moved a centimetre for about 130 µV. A galvanometer is driven by current, so the electrodes sit in series with its coil and every ohm of contact costs deflection. A good deal of the work was therefore electrode work: clay gave way to metal, contact resistances as high as 44 kΩ came down to a few hundred ohms, and for preference he pushed needles through the scalp to the periosteum.

An amplifier ends that dependence, because it senses voltage and draws almost no current. Berger himself was using a Siemens amplifier by 1931. When Edgar Adrian and Bryan Matthews repeated his experiments at Cambridge in 1934, their valve amplifier took next to nothing from the electrodes, so skin resistance no longer set the size of the waves, and squares of copper gauze soaked in warm saline and bandaged to the head did as well as needles. Their usual first stage was one Matthews had described that March: a balanced pair of valves with neither input at earth, so that interference raising and lowering both electrodes together relative to earth left the output unchanged. That is a differential amplifier, and it belongs to the confirmation, not the discovery; by 1938 Otto Schmitt and J. F. Toennies had both published circuits under that name.

Modern EEG amplifiers are differential too, and typically reject what the two inputs share by more than 100 dB, a factor of 10⁵ — a figure that holds only if the electrodes match. Interference on the body reaches each input through a divider formed by that electrode’s impedance and the amplifier’s input impedance, so unequal electrodes divide it unequally and part of it arrives as a difference, roughly the mismatch over the input impedance. Two electrodes that differ by 10 kΩ, into a 100 MΩ input, turn one part in 10⁴ of the interference into signal: 80 dB of rejection at best, whatever the amplifier manages on its own. So scalp sites are still abraded and gelled, taking contact impedance from 150–200 kΩ down to 5–10 kΩ. The amplifier did not end Berger’s electrode problem; it turned lost signal into admitted interference.

What it established

That the brain is always electrically active, in rhythms, and that those rhythms change with state — eyes open or closed, awake or asleep, seizing or not. Before this, cortical electrophysiology was mostly about evoked responses to stimuli. EEG made spontaneous activity a subject.

It took five years to be believed. In 1934 Adrian and Matthews at Cambridge and, independently, Hallowell Davis at Harvard confirmed it. Adrian and Matthews had set out doubting that activity so uniform could run through a waking brain, and they confirmed the rhythm but not Berger’s view of its source: about ten waves a second, present with the eyes shut and abolished by any attempt to see pattern, and arising in the occipital lobes rather than the whole cortex. They named it the Berger rhythm, drove it at the rate of a flickering light, and found its frequency all but fixed — set, they argued, by the properties of the cells and nothing else. Adrian produced it whenever he closed his eyes; in Matthews it came and went, and he made the better observer.

Its clinical life followed immediately: epilepsy diagnosis, sleep staging, and depth-of-anaesthesia monitoring all rest on reading these rhythms, and all of them predate any mechanistic account of what generates them.

In 1935 Gibbs, Davis and Lennox reported spike-and-wave discharges at three a second during absence seizures; Berger had recorded the same pattern in 1931 but did not show it until 1938, suspecting blinks and facial movements. The same year Alfred Loomis, Newton Harvey and Garret Hobart recorded sleepers through the night on a drum eight feet long, and called the bursts at 14 a second that swelled and faded spindles, after their shape on the record. By 1957 William Dement and Nathaniel Kleitman were scoring sleep in four EEG stages, from stage 1, without spindles, to stage 4, with at least half the record in slow waves of 100 µV or more. Every period of rapid eye movements they recorded fell in stage 1, and the stages cycled regularly through the night. A sleep stage, in that scheme, is a pattern in the EEG.

Patterns are only comparable if the electrodes are. Herbert Jasper’s 10–20 system, published for the International Federation in 1958, puts 21 electrodes at 10% and 20% steps of distances measured between landmarks on the skull, so the same layout and the same names scale to any head.

The trade it makes

EEG has excellent time resolution and poor spatial resolution, and the reason is physical rather than technological. The skull is a low-conductivity layer between source and sensor, so potentials spread and superpose; what reaches an electrode is a blurred sum over a large volume of synchronously active cortex. Solving backwards from scalp voltages to sources is ill-posed — many source configurations produce the same measurement.

That limitation is the mirror image of fMRI, which has the spatial resolution and lacks the timing, and it is why the two are so often used together.

Because the signal requires synchrony across many neurons to be visible at all, EEG is inherently a population measurement: activity that is not coordinated cancels before it reaches the scalp.

The currents that survive are mostly synaptic. An action potential lasts a millisecond or two, and neighbouring neurons rarely fire within so short a window, so their fields seldom add; synaptic currents are slower, and in cortex the apical dendrites of pyramidal cells lie parallel, so their dipoles can sum. In Buzsáki, Anastassiou and Koch’s account, one scalp electrode sees the local field averaged over 10 cm² or more of cortex, and usually says little about how the individual neurons beneath it are firing. The magnetic counterpart, magnetoencephalography, picks up the same currents through their magnetic fields, and those depend much less on the conductivity of the tissue around them.

Adrian and Matthews had named the cost in 1934. A rhythm large enough to detect through the skull needs a large area of cortex beating in step, which is unlikely anywhere busy handling the messages coming in and going out; on their reading the Berger rhythm “shows what happens in an area of cortex which has nothing to do”. Recordings through the intact skull, they concluded, would seldom show what particular regions were doing in their normal work. Better amplifiers have not changed that argument, because it is about the source, not the instrument.

The standard layout has blind spots as well. Its positions do not cover the front and underside of the temporal lobes, the commonest source of epileptic activity, which is why the IFCN’s 2017 guideline added an inferior temporal chain for a basic array of 25 electrodes. High-density caps of 64–256 electrodes, with source modelling, localise below the scale of a lobe. Single neurons stay out of reach from the scalp; for those the electrodes go into the tissue, as with microelectrode arrays.