Why magnetic, why on-scalp
Magnetoencephalography measures the minute magnetic field produced by synchronised neuronal currents. Because tissue is magnetically nearly transparent, MEG suffers far less of the skull-induced blurring that plagues EEG — its inverse problem is better conditioned. The price: the fields are tiny, on the order of femto- to pico-tesla, roughly a billionth of the Earth's steady field.
|\mathbf{B}(\mathbf{r})| = \frac{\mu_{0}}{4\pi}\,\frac{|\mathbf{Q}|\,\sin\alpha}{r^{2}}The field of a neuronal current dipole grows steeply as the sensor nears the source. Halving the sensor-to-brain distance is the entire economic case for putting magnetometers on the scalp.
The magnetic field from a neural current dipole also drops with the square of distance r, so getting the sensor closer wins big — halve the distance and the signal roughly quadruples. That single fact is the whole argument for moving magnetometers off a bulky helmet and onto the scalp.
- |\mathbf{B}(\mathbf{r})|
- The magnetic field strength at the sensor.
- |\mathbf{Q}|
- The strength of the neuronal current dipole.
- r
- Sensor-to-source distance; the 1/r^2 falloff is the key.
- \sin\alpha
- Orientation factor — a dipole pointing straight at the sensor gives no field.
A traditional MEG helmet sits about 4 cm from cortex; an on-scalp OPM can sit about 1.5 cm away, and the 1/r^2 law turns that into several-fold more signal.
Conventional MEG uses superconducting SQUIDs that must be bathed in liquid helium, forcing a ~2 cm thermal gap between sensor and scalp — and a fixed, one-size helmet. An optically pumped magnetometer runs warm, so it can sit directly on the head. Closing that gap raises the measured signal several-fold and sharpens spatial resolution, precisely because of the steep falloff above. This is on-scalp MEG.
Inside an OPM: SERF magnetometry
An OPM is a millimetre-scale glass cell of alkali vapour (rubidium or caesium). A pump laser optically polarises the atomic spins; an external magnetic field makes those spins precess; a probe beam reads the resulting change in transmitted light. In the spin-exchange relaxation-free (SERF) regime — near-zero field and high atomic density — spin-destroying collisions are suppressed, coherence times lengthen, and sensitivity soars into the femtotesla range.
\delta B = \frac{1}{\gamma}\,\frac{1}{\sqrt{n\,T_{2}\,V\,t}}Spin-projection-noise limit of an atomic magnetometer: sensitivity improves with atomic density n, coherence time T₂, cell volume V and averaging time t. These are exactly the SERF knobs — and they explain the fundamental tension with miniaturisation (small V hurts).
The best an atomic magnetometer can do is limited by the random "noise" of the atoms it uses to sense. This says the noise floor shrinks — sensitivity improves — when you have more atoms (n), let them stay coherent longer (T_2), use a bigger cell (V), or average longer (t). The catch: shrinking the cell to fit on the scalp lowers V and hurts you.
- \delta B
- The smallest magnetic-field change the sensor can resolve (its noise floor).
- n
- Atomic density — more atoms average the noise away.
- T_2
- Coherence time — how long the atoms stay in step.
- V
- The volume of the vapor cell.
- t
- Averaging time — a longer measurement means less noise.
Everything under the square root helps only as a fourth-root of \delta B: to halve the noise floor you must raise the product n\,T_2\,V\,t sixteenfold — which is why miniaturising the cell is so costly.
The wearability payoff — and its catch
Because OPMs mount on a lightweight cap that moves with the head, they unlock paradigms a rigid SQUID helmet forbids: naturalistic movement, children and developmental cohorts, longer recordings — all with intrinsically larger signals. The new hard problems are engineering ones: sensor cross-talk, precise calibration and co-registration of many small sensors, and above all suppression of the environmental and self-generated background field.
Interference suppression borrows from classical MEG — projection methods such as signal-space separation / Maxwell filtering — but the assumptions shift when sensors sit close and inside the source-free-shell geometry is different. And OPM data still feed the full Volume II source-localisation stack (forward model, then a regularised inverse); the lead fields simply differ because the sensors are nearer and on-scalp.
Emerging: solid-state quantum magnetometry
Beyond vapour cells, nitrogen-vacancy (NV) diamond magnetometers promise room-temperature, potentially unshielded, solid-state sensing at high spatial resolution. Today their sensitivity trails OPMs by orders of magnitude, so they are not yet competitive for scalp MEG — but they exemplify the broader acousto-opto-magnetic sensing frontier, where a physical transducer is swapped for a quantum one and the whole tradeoff re-opens.