Room-temperature, on-scalp magnetoencephalography — MEG timing and localization without the cryogenics or the fixed helmet.
An optically pumped magnetometer measures a magnetic field by polarising a vapour of rubidium, caesium, or potassium atoms with laser light and reading the change in their polarisation. Nothing needs cooling to a few kelvin. The first convincing on-scalp evoked-field recordings appeared in 2016, and the technology has since moved from a physics curiosity to a shipping product line.
The critical property is where the sensor sits. A SQUID MEG system measures from a fixed helmet position several centimetres outside the scalp; an OPM array places sensors directly on the scalp. Since a dipolar field falls off as the inverse cube of distance, moving from helmet to scalp is a large SNR gain rather than a detail — a sensor 1 cm from the cortex sees far more of the field than one 4 cm away.
Cryogenic MEG requires matching a rigid, heavy dewar to each subject’s head, and the head-shape template used in the forward model is a significant error source. OPM arrays are constructed on the head itself: sensors on a flexible, custom-fitted frame, individually positioned, with individual digitisation folded into the measurement rather than estimated afterwards.
A consequence people miss: the array stays on the head during the run, so the subject can move — head rotations, natural posture, even a bit of fidgeting. Cryogenic systems require a dewar that tracks the head within millimetres, which is precisely what a participant cannot guarantee. This is what makes OPM-MEG viable in paediatric cohorts and in movement experiments.
The most valuable consequence of on-scalp is genuine co-registration. A modular helmet carries MEG gradiometers and dense EEG electrodes in one fixture, so both are sampled by the same hardware, digitised in the same coordinate frame, and cannot drift relative to each other. Combined MEG/EEG reconstruction becomes a well-posed measurement rather than a cross-modal registration exercise.
The sensor density follows the same logic. Fit for a child’s head at a few centimetres spacing, the same sensor can be repositioned and regapped for an adult — 300+ channels over a full adult scalp is now routine. Recording density is a data variable, not a hardware decision made years in advance.
OPMs do not make shielding optional in the way the hype suggests. Their sensitivity to field is comparable to good SQUID systems, so ambient field at the tens of microtesla level is still the binding constraint. Three responses are in use: a passive shield (heavy copper or mu-metal enclosure), an active compensation field generated inside the room, or gradiometric cancellation — measuring the difference between a sensor and a nearby reference to reject common-mode field, which lets a compact portable shield suffice for many applications.
Open questions remain around noise as a function of head motion (field gradients across a non-rigid sensor array break the common-mode assumption), sensor saturation near strong sources, and long-term drift of the optical readout. Good pipelines handle motion correction and field prediction explicitly; treating OPM as a drop-in SQUID replacement is how artefacts get into the literature.