Quantum Sensing in MRI: How SQUID Magnetometers Work (2026)

Brains and hearts radiate magnetic fields about a hundred million times weaker than the Earth’s. Picking those signals up has always required exotic detectors, and in 2026 the lineup of those detectors is changing fast. Quantum sensing MRI SQUID magnetometer technology — once a niche of low-temperature physics labs — now anchors a generation of brain-imaging systems that sit between classical MRI and clinical electroencephalography. Optically pumped magnetometers and nitrogen-vacancy diamond sensors are joining the SQUID, each with its own thermal, geometric, and noise-floor trade-offs. This research explainer walks through what these sensors actually measure, how they compare, and where the engineering still hurts.

What this post covers: the physics of SQUIDs, OPMs, and NV centers; how each one slots into MRI and magnetoencephalography pipelines; current clinical-pilot evidence from 2025 and 2026; the shielding, cryogenics, and dynamic-range problems that remain; and where low-field portable scanners may plausibly land.

From classical MRI to quantum-sensed MRI

Answer-first summary: Classical MRI detects nuclear precession by inducing a voltage in a room-temperature radio-frequency coil, and it depends on a strong static field of 1.5 to 7 tesla. Quantum-sensed MRI and magnetoencephalography flip that around. They use ultra-sensitive magnetometers — typically SQUIDs, optically pumped magnetometers, or NV-center diamonds — to read tiny magnetic fields directly. The static field can be far weaker, or absent.

Classical clinical MRI exists because protons in the body precess at a Larmor frequency that scales with the applied static field. A 1.5 T scanner places protons at roughly 64 MHz; a 3 T scanner doubles that. An RF coil picks up the relaxing magnetization, an analog front-end samples it, and a reconstruction stage maps the k-space samples into an anatomical image. This is well-trodden ground, and its sensitivity comes from sheer field strength rather than detector finesse.

Quantum-sensed systems take the opposite path. The detectors themselves are exquisite. SQUIDs reach the femtotesla regime; commercial optically pumped magnetometers report similar noise floors in tuned configurations; and some NV-center setups now claim picotesla-class sensitivity at room temperature. With detectors that good, you no longer need brute-force fields. You can use microtesla bias fields, hyperpolarization, or no static field at all — and still pull out useful signals. You can also leave the body alone and listen to the magnetic chatter the brain emits on its own.

Two distinct application families fall out of this. The first is low-field MRI, where a quantum sensor reads nuclear precession after a brief pre-polarizing pulse. The second is magnetoencephalography, or MEG, where the sensor reads neural currents directly without any applied RF.

If you have not seen comparable physics-led measurement shifts elsewhere, our cryo-EM 1.2 angstrom resolution milestone explainer covers another field where detector improvement, not magnet strength, unlocked the next decade. The pattern is similar.

What SQUIDs actually measure: flux quanta and Josephson junctions

Answer-first summary: A SQUID — superconducting quantum interference device — measures magnetic flux threading a small superconducting loop interrupted by one or two Josephson junctions. The output voltage oscillates with each flux quantum that crosses the loop. A feedback loop linearizes the response, giving a clean measurement of magnetic field or its rate of change in the femtotesla per root hertz range, depending on geometry and bias.

The physics is the most counter-intuitive of the three sensors we will cover, so it is worth being precise. A superconducting loop refuses to carry continuous magnetic flux. Instead, flux through the loop is quantized in units of the flux quantum, phi-zero, equal to Planck’s constant divided by twice the electron charge — about 2.07 times ten to the minus fifteen webers. When you place one or two Josephson junctions in the loop, you create a tunnel barrier whose current-phase relationship is sensitive to the enclosed flux.

In the DC SQUID, two junctions in parallel produce an interference pattern. As external flux changes, the critical current and the voltage across the junctions oscillate periodically with phi-zero. The slope of that oscillation is steep. A small change in field translates into a measurable voltage change.

Real SQUIDs do not run open-loop. They are wrapped in a flux-locked loop, or FLL, that injects a counter-flux through a feedback coil to keep the SQUID at its most sensitive operating point. The amplifier output then becomes a linear function of the applied field, often calibrated to give an output in tesla.

Published noise floors for biomedical SQUIDs typically sit in the low single-digit femtotesla per root hertz at low frequencies, with several groups reporting performance approaching one femtotesla per root hertz under favorable shielding. Treat any specific number as instrument-specific and bandwidth-specific; the underlying figure depends on junction noise, coupling efficiency, and external interference.

Why SQUIDs sit in liquid helium

Low-temperature SQUIDs use niobium-based junctions and need to be around 4.2 kelvin. That means a liquid-helium dewar surrounding the sensor head. High-temperature SQUIDs based on YBCO ceramics operate around 77 kelvin in liquid nitrogen, which is cheaper, but their intrinsic noise tends to be higher. For brain imaging, low-temperature SQUIDs dominate clinical MEG installations, while high-temperature SQUIDs appear more in research and industrial sensing.

What SQUIDs do not do well

SQUIDs are flux sensors, not absolute field sensors. They can lose their lock if the field changes too quickly. They demand careful magnetic shielding. And their dewar geometry forces a fixed sensor-to-scalp gap, often two to three centimetres, which dilutes spatial resolution for MEG.

How an OPM compares: optically pumped magnetometers

Answer-first summary: An optically pumped magnetometer uses laser-polarized atomic vapor, typically rubidium or potassium, as the sensing element. The Larmor precession of those atoms shifts in proportion to the external magnetic field, and that shift modulates the light passing through the cell. OPMs hit noise floors comparable to SQUIDs in the right operating regime, but they run near room temperature, allowing scalp-contact MEG helmets.

OPMs replace the cold superconductor with a centimetre-scale vapor cell. A pump laser, often tuned to the D1 line of rubidium near 795 nm or potassium near 770 nm, polarizes the electron spins of the atoms. In zero or near-zero ambient field, the spins precess at a frequency proportional to the residual field. A probe beam reads that precession via optical rotation or absorption. A lock-in amplifier extracts the field.

The class of OPM that has dominated 2024 to 2026 MEG work is the spin-exchange relaxation-free, or SERF, magnetometer. SERF cells sit at high alkali density and elevated temperature (typically 150 to 180 degrees Celsius) so that spin-exchange collisions no longer broaden the resonance. In that regime, the cell can reach extremely low noise floors — vendor specifications and peer-reviewed measurements often cluster in the tens of femtotesla per root hertz, with some research-grade demonstrations going lower.

The decisive practical advantage is geometry. Because the cell does not need cryogenic isolation, it can sit a few millimetres from the scalp, inside a soft, child-friendly or adult-friendly helmet. That proximity raises signal strength compared to the fixed two-to-three-centimetre gap typical of dewar-based SQUID arrays. Several vendors, including Cerca Magnetics and QuSpin, have pushed wearable OPM-MEG systems into clinical and research deployments. Independent technical briefs from those vendors document the engineering choices, and peer-reviewed validation has progressed since the 2018 Nature paper by Boto and colleagues that first demonstrated wearable OPM-MEG.

Trade-offs of OPMs

OPMs work well in a narrow bandwidth around zero ambient field, which is why they live inside heavily shielded rooms. Their dynamic range is small — typically a few nanotesla — so motion through residual gradients can saturate them. Modern systems compensate with on-helmet bias coils and active field nulling, but the engineering is non-trivial.

NV-center diamond magnetometry: room-temperature quantum sensing

Answer-first summary: A nitrogen-vacancy, or NV, center is a point defect in diamond consisting of a substitutional nitrogen atom next to a missing carbon. The defect has a spin-triplet ground state whose sub-levels split under a magnetic field. By optically pumping the spin and reading its fluorescence under microwave driving, you can measure magnetic field. NV magnetometers run at room temperature, can be miniaturized, and offer spatial resolution that is hard to match with SQUIDs or OPMs.

NV centers have a green-pumped, red-emitting fluorescence cycle that preferentially polarizes the electron spin into one of three sub-levels of its triplet ground state. A microwave drive at roughly 2.87 gigahertz addresses the transition between those sub-levels. An external magnetic field Zeeman-splits the resonances. By sweeping the microwave frequency and watching the fluorescence dip, you read out the field. The technique is called optically detected magnetic resonance, or ODMR.

Three properties make NV magnetometry attractive for biomedical sensing.

• Room temperature operation. No cryogenics, no high-temperature vapor cells.
• Spatial resolution. A single NV center is atomic in scale. Ensembles in diamond chips can reach micrometre or sub-micrometre voxels.
• DC capability. Many configurations sense absolute static fields, not only AC or transients.

Reported sensitivities for biomedical-grade NV magnetometers vary widely. Bulk diamond sensors using laser-threshold or pulsed-readout techniques have demonstrated sub-picotesla per root hertz performance in laboratory conditions, with some groups projecting femtotesla regimes after further engineering. Many production-ready devices today sit in the picotesla range. Treat all such numbers cautiously, and read the original methods sections — sensitivity depends on integration time, ODMR contrast, and photon collection efficiency.

The Quantum Diamond Technologies group, the Fraunhofer institutes, and several MIT spin-outs have published roadmaps that mix academic and engineering milestones for clinical NV magnetometry. The trajectory is real, but the timeline is hedged for good reason.

Where NV centers help and where they do not

NV magnetometers shine in tight-geometry sensing — small animal MEG, magnetocardiography of single hearts in research, and emerging current-imaging tasks in semiconductor inspection. They are not yet a drop-in for adult-head MEG arrays that need 64 to 256 channels at comparable sensitivity to OPMs. That gap is closing, but as of 2026 it has not closed.

Magnetoencephalography clinical pilots, 2025 to 2026

Answer-first summary: Magnetoencephalography is the clearest near-term clinical home for these quantum sensors. SQUID-based MEG has been an approved adjunct for epilepsy surgery planning for years. OPM-MEG entered clinical pilot trials around 2023 to 2024 and continues to expand. NV-center MEG remains primarily research-grade. The clinical evidence base is growing, but each modality still has well-defined limits.

Adult and pediatric epilepsy surgery is the application where MEG has the strongest accepted clinical role. SQUID-based MEG installations at major academic centres routinely localize interictal spikes to support surgical planning. The technology has been mature for two decades, and reimbursement pathways exist in several health systems, although they vary by jurisdiction.

OPM-MEG raises two specific clinical promises. First, paediatric brain imaging becomes much easier with a soft, light helmet that scales to a child’s head and tolerates motion better than a fixed dewar. Second, functional brain mapping in awake, moving patients opens tasks that SQUID-MEG cannot do — for example, gait studies, naturalistic visual paradigms, and speech production under realistic conditions. Peer-reviewed clinical pilots since 2023 have demonstrated OPM-MEG concordance with established SQUID-MEG benchmarks on standard evoked-response paradigms.

Functional brain SQUID 2026 deployments still anchor the high end of clinical MEG, but vendor and academic investment has clearly shifted toward OPM systems for new builds. Both modalities co-exist in research labs, and several centres run head-to-head comparisons.

A few caveats are essential to keep on the table:

• Clinical efficacy claims should be cited from peer-reviewed clinical studies, not vendor brochures. Brochures describe technical specifications; brochures do not establish clinical utility.
• MEG, in any flavor, is an adjunct technology. It complements EEG and MRI rather than replacing them.
• Generalization from epilepsy localization to broader psychiatric or neurological diagnostics is still active research, not standard care.

Engineering challenges: shielding, cryogenics, and dynamic range

Answer-first summary: Three engineering domains determine whether a quantum-sensed brain imaging system works in a real building. Shielding suppresses ambient magnetic interference by four to six orders of magnitude. Cryogenics dictates dewar geometry and helium logistics for SQUIDs. Dynamic range determines whether motion and gradients drive the sensor out of its linear regime. Each domain has well-developed engineering, but each also imposes hard physical limits.

Magnetic shielding

Brain magnetic fields range from roughly tens of femtotesla for evoked responses to a few picotesla for spontaneous alpha waves. The Earth’s field, by contrast, is about 50 microtesla, and a passing elevator can swing local fields by tens of nanotesla. Useful brain measurements therefore require shielding that suppresses ambient field by a factor of ten thousand to a million.

That is the job of a magnetically shielded room, or MSR. A modern MSR layers high-permeability mu-metal with aluminium for eddy-current shielding. Active nulling coils trim residual DC components and slow drifts. Some OPM-MEG installations use lighter active-only shielding combined with on-helmet field control, trading some attenuation for cost and footprint. The economics depend on the building, not just the sensor.

Cryogenics

SQUID-MEG systems live and die by their helium logistics. A typical dewar consumes tens of litres of liquid helium per week unless equipped with a closed-cycle re-condenser. Helium price spikes — and global supply has had several — directly hit operating cost. Re-condensing dewars cut consumption sharply but raise capital expense and add a continuous compressor noise source that itself must be magnetically managed.

High-temperature SQUIDs running on liquid nitrogen sidestep the helium problem but trade sensitivity. OPMs and NV magnetometers sidestep cryogenics entirely.

Dynamic range and motion

A SERF OPM saturates outside a window of a few nanotesla. Move a sensor through a residual Earth-field gradient of even a few nanotesla per centimetre and it falls out of its operating regime. Active on-helmet coils and reference sensors compensate, but the engineering is involved. SQUIDs handle larger fields but can still slip a flux lock, which then requires reset and a brief loss of data continuity. NV magnetometers can have much larger dynamic range, but at the cost of sensitivity in their straightforward configurations.

Where this is heading: low-field MRI and portable scanners

Answer-first summary: Quantum sensors enable two specific directions of MRI evolution: ultra-low-field MRI for portable, point-of-care imaging, and quantum-sensed magnetic resonance modalities that combine pre-polarization with magnetometer readout. Neither has displaced 1.5 T or 3 T MRI for general diagnostic work, and neither is likely to in this decade. But both have niches where they may dominate.

The portable low-field scanner market — exemplified by Hyperfine’s 64 mT system — already exists with conventional RF coils. Quantum-sensed approaches push the field down further. With a SQUID or OPM reading the precession signal, useful images can come from millitesla or even microtesla static fields after pre-polarization. The image quality at those fields will not match a hospital 3 T magnet for soft-tissue contrast, but the form factor and cost can be dramatically different.

A few realistic application targets:

• Point-of-care neurological triage, particularly in stroke or trauma settings, where speed and accessibility matter more than ultimate resolution.
• Hyperpolarized agent imaging, where the dominant signal source is no longer thermal Boltzmann polarization but rather an external preparation, decoupling MRI sensitivity from the static field.
• Functional brain mapping combining low-field MRI with co-located MEG, leveraging the same sensor array for both modalities.

This direction parallels how high-precision timing went from lab atomic standards to portable systems — see our explainer on how atomic clocks work and the related precision timekeeping piece on GPS atomic clocks for context on how that miniaturization unfolded.

The clinical-pilot data flow for any of these systems is structurally similar across modalities. Sensors feed an analog front-end, a digital acquisition layer, an artifact-rejection and inverse-modelling pipeline, and finally a clinical visualization layer.

The processing side, increasingly, is where machine learning enters. Source localization and artifact rejection benefit from learned priors, and several centres are exploring privacy-preserving multi-site training. If that interests you, the federated learning IoT architecture article covers the same FedAvg and FedProx patterns that some MEG and low-field MRI research consortia are starting to use.

Trade-offs and gotchas

Answer-first summary: Every quantum sensor for biomedical imaging trades sensitivity for geometry, geometry for thermal management, and thermal management for cost. SQUIDs deliver the most consistent femtotesla performance but lock you into helium logistics. OPMs offer wearable form factors but a narrow operating window. NV centers shine at room temperature with tight spatial resolution but trail SQUIDs and OPMs in adult-MEG sensitivity. Pick by application, not by hype.

A short list of recurring traps:

• Reported noise floors are bandwidth-specific. A femtotesla per root hertz figure at 10 to 100 Hz is not the same as a figure at 1 Hz. Always read the bandwidth.
• Sensitivity in a shielded room is not the same as sensitivity in your hospital. Local interference dominates many real installs.
• OPM “wearable” does not mean motion-tolerant by default. Many published demonstrations include motion limits, on-helmet field nulling, and reference sensors that are not always emphasized in marketing.
• NV magnetometry timelines are honest but lengthy. Several groups have demonstrated remarkable single-NV performance; scaling to clinical helmet-style arrays is a separate engineering problem.
• Clinical evidence is per-application. SQUID-MEG for epilepsy is well-supported; the same SQUID-MEG for psychiatric biomarkers is not — that is research.

Beyond these, the broader gotcha is interpretive. Quantum sensors give you better magnetometers. They do not, by themselves, give you better diagnosis. The forward and inverse problems — turning measured magnetic field maps into cortical activations — remain hard and assumption-laden. A sharper sensor amplifies the consequences of bad co-registration, bad head models, and bad source priors.

Practical recommendations for biomedical engineers

Answer-first summary: If you are scoping a quantum-sensing MRI or MEG project in 2026, anchor your decisions in three steps: clarify the clinical or research question, pick the sensor modality that fits the geometry and motion profile of your target population, then size shielding and processing around the chosen sensor. Build in measurement protocols that let you compare against the established SQUID-MEG baseline.

A working checklist for project planning:

• Define your acceptable spatial resolution, temporal resolution, and target population. Adult, paediatric, and neonatal MEG have different geometric constraints.
• Survey the building. Magnetically shielded room costs and feasibility dominate total cost of ownership for any femtotesla-class system.
• For SQUID systems, plan helium logistics or budget for a re-condensing dewar.
• For OPM systems, plan on-helmet field nulling, reference sensors, and a controlled motion envelope.
• For NV-center pilots, treat them as research instruments unless you have direct validation in your application.
• Run a SQUID-MEG comparison whenever possible. Even one or two head-to-head sessions catch most systematic biases early.
• Use open processing toolchains — MNE-Python, FieldTrip — and publish your forward and inverse model parameters.

Quantum biomagnetism, as a field, is at the same point AI-driven structural biology was several years ago: real capability, real limits, and a strong incentive to read the methods sections carefully. For an adjacent example of where AI now competes with classical methods on a real biological problem, our AlphaProteo de novo protein binder design architecture piece is a useful counterpoint.

FAQ

Is a SQUID magnetometer the same as an MRI scanner?

No. A SQUID is a magnetic-field sensor, not an imaging system on its own. Classical clinical MRI uses radio-frequency coils, not SQUIDs. Some research-grade and emerging low-field MRI systems use SQUIDs to detect nuclear precession in microtesla fields. SQUIDs are also the workhorse sensor in magnetoencephalography, which images brain currents rather than tissue structure. The terminology blends because both modalities measure magnetic fields, but the imaging principles differ.

How does OPM-MEG compare to SQUID-MEG in clinical use?

OPM-MEG offers a wearable helmet, scalp-proximate sensors, and motion tolerance that SQUID-MEG cannot easily match. Reported on-scalp signal strengths are higher for OPM systems because the sensor-to-cortex distance is smaller. SQUID-MEG, however, has a longer clinical track record, particularly for epilepsy surgery planning, and most clinical reimbursement frameworks reference SQUID systems. OPM-MEG clinical pilots since 2023 show concordance on standard paradigms, but they are not yet a one-to-one replacement in every clinical workflow.

What sensitivity do NV-center magnetometers reach?

It depends on the configuration and the bandwidth. Bulk-diamond NV ensembles using pulsed and laser-threshold techniques have reported sub-picotesla per root hertz sensitivities in laboratory settings, with theoretical and demonstrated paths toward femtotesla regimes. Many production-ready devices currently sit in the picotesla range. Always check the integration time, the optical readout contrast, and the photon collection efficiency in the original publication. Quoted noise floors without those parameters are not directly comparable across groups or vendors.

Do quantum-sensed MRI scanners replace 3 tesla machines?

No, not in 2026 and almost certainly not within this decade. High-field MRI dominates soft-tissue diagnostic imaging because of its raw signal-to-noise and contrast options. Quantum-sensed low-field systems target different niches — point-of-care imaging, paediatric and neonatal use, combined MEG-MRI workflows, and hyperpolarized agent imaging. They are likely to coexist with high-field scanners, addressing applications where portability, cost, or functional readout matter more than peak anatomical resolution.

What kind of shielding does a quantum-sensing MEG lab need?

Most installations use a magnetically shielded room built from layered mu-metal and aluminium, often with active nulling coils. Attenuation requirements depend on local environment but typically span four to six orders of magnitude across the bandwidth of interest. OPM-MEG systems sometimes pair lighter passive shielding with aggressive on-helmet active control, trading attenuation for footprint and cost. The decision is site-specific and should follow a measured ambient noise survey rather than catalogue defaults.

Are NV-center magnetometers safe to use on patients?

The diamond sensor itself is inert and sits external to the body, so direct material safety is not a concern. The optical and microwave subsystems are engineered to standard biomedical safety levels, with laser power and microwave exposure managed by the device design. As with any clinical instrument, regulatory approval is application-specific. As of 2026 most clinical NV-center magnetometry remains in research contexts. Assume any device targeting patient use will require region-specific regulatory clearance.

Further reading

• Cryo-EM 1.2 angstrom resolution milestone explained — another field where detector improvement, not raw signal, unlocked progress.
• How atomic clocks work: quantum resonance and GPS — quantum sensing in the time-frequency domain.
• How GPS atomic clocks work: precision timekeeping — companion piece on the engineering of portable quantum standards.
• Federated learning IoT architecture with FedAvg and FedProx — privacy-preserving multi-site learning patterns now appearing in MEG research consortia.
• AlphaProteo de novo protein binder design architecture — adjacent example of measured-versus-hyped capability in biomedical AI.

External references worth reading in full:

• Clarke, J. and Braginski, A. I., editors, The SQUID Handbook, Wiley-VCH — the standard SQUID physics and engineering reference.
• Fagaly, R. L., “Superconducting quantum interference device instruments and applications,” Review of Scientific Instruments, on SQUID design and noise.
• IEEE Transactions on Applied Superconductivity — ongoing peer-reviewed venue for SQUID and superconducting-electronics engineering.
• Nature Physics reviews on SQUID magnetometry and on quantum sensing — for the cross-domain physics framing.
• Boto et al., “Moving magnetoencephalography towards real-world applications with a wearable system,” Nature, 2018 — the foundational OPM-MEG demonstration.
• Brookes et al. and subsequent OPM-MEG clinical pilot papers, 2023 to 2025, in NeuroImage, Brain, and related journals.
• Cerca Magnetics and QuSpin technical briefs on commercial OPM-MEG systems — useful for engineering specifications.
• Barry et al., “Sensitivity optimization for NV-diamond magnetometry,” Reviews of Modern Physics, on the sensitivity-budget side of NV magnetometry.
• Fraunhofer IAF and MIT Quantum Engineering Group white papers on NV-center diamond magnetometry — for current engineering roadmaps.
• Quantum Diamond Technologies Inc. publications on biomedical NV-center applications.

---

Riju writes about industrial IoT, digital twins, PLM, and adjacent applied-science topics on iotdigitaltwinplm.com.