The Next Stage in Magnetoencephalography
Widely accessible brain-computer interfaces will require improvements in non-implanted technology. With advances in quantum sensing, MEG has enormous potential.
In an age of accelerating AI, e184 is working to preserve a human stake in the future. We want technology that can achieve human-AI symbiosis, via a brain-computer interface powerful and practical enough to benefit people around the world. Over the last year, we assessed the field, characterizing the capabilities we need and technologies with the potential to get there. We determined that we would need a technology that doesn’t require an implant, that can reach spatial resolutions of a millimeter and temporal resolutions of tens of milliseconds. With those criteria, we have a clear first choice: magnetoencephalography.
Magnetoencephalography, or MEG, measures magnetic fields produced by coordinated currents in dendrites of pyramidal neurons. When currents in the cerebral cortex are lined up tangentially with the skull, they lead to measurable, if small, fields outside the head. These fields are on the order of femto-Tesla, much lower than the Earth’s magnetic field, which is around fifty micro-Tesla.
Measuring these tiny fields is the fundamental challenge of MEG. The approach has evolved dramatically over the years from David Cohen’s first experiments with copper coils in 1968, and is now able to reach greater sensitivity in more varying conditions. We plan to take it even further.
SQUIDs: MEG 1.0
While Cohen’s first experiment used an induction coil, he soon switched to what at the time was a brand-new technology, the superconducting quantum interference device, or SQUID. To date, SQUIDs still form the basis for the large majority of MEG devices in practical use. They can detect extremely tiny magnetic fields, down to several atto-Tesla in ideal cases, making them capable of registering the extremely tiny fields generated by the brain.
Due to their use of superconductors, SQUID MEG sensors must be contained in dewars of liquid helium. Hundreds of sensors are typically arranged in arrays to achieve quick readout from different parts of the brain, with each sensor registering signals with a single orientation. The volume of cryogenic equipment needed positions these arrays at around 2cm from the subject’s head. The result is a bulky, rigid setup, one that requires patients to sit still and have a head of roughly the right size and shape, making it difficult to impossible to use on children.
The SQUIDs used in MEG have a dynamic range of twenty nano-Tesla, high for a magnetometer but still too low to span the ten orders of magnitude between signals of interest and the Earth’s magnetic field. As such, all SQUID MEG is performed in magnetically shielded rooms, sophisticated installations with layered metal to filter out ambient sources of magnetic noise. These rooms are a feasible investment for hospitals and universities, but obviously not viable for a device meant for everyday use.
OPMs: MEG 2.0
SQUIDs use loops of superconducting material to create systems that are extremely sensitive to magnetic fields. Over time, it has become increasingly feasible and practical to create atomic and solid-state systems with this kind of sensitivity, a research field broadly referred to as “quantum sensing”. In recent years, quantum sensing approaches have leapfrogged ahead, finding advantages in a variety of settings from astronomy to biology.
Optically pumped magnetometers, or OPMs, are a quantum sensing technology that has made remarkable strides in magnetoencephalography. By using light to “pump” atoms into particular quantum states, OPMs control the spin state of atoms in vapors of alkali metals like rubidium, creating a system that is highly sensitive to magnetic fields without relying on the cryogenic cooling needed by superconducting systems.
The 1957 observation that optical pumping can create a magnetically sensitive state occurred even earlier than the prediction of the Josephson effect that lies behind SQUIDs, and by the 1970’s researchers had dramatically improved sensitivity and started the path towards miniaturization. Still, it took almost thirty years for the technology to become competitive with SQUIDs for applications that required compact sensors, with a resurgence of interest driven by the surprising observation that high-density vapors can avoid reductions in sensitivity caused by colliding atoms exchanging spin. The discovery of this spin exchange relaxation free (SERF) regime led to a expansion of neuroscience applications of OPM technology, starting in the 2010’s and continuing to today.
With these sensitivity gains, OPMs’ lack of need for cryogenics constitutes a potential massive advantage over SQUID systems. It means OPM MEG sensors can be placed much closer to a patient’s scalp, and can move flexibly to account for heads of different shapes, for a three-to-five-fold improvement in sensitivity. The systems can be made mobile, with comparably light headsets linked by cables to an apparatus comparable to a large backpack, allowing for MEG studies of subject in motion, as well as medical uses on children and others who would have trouble sitting still enough for SQUID-based MEG. OPMs are also vector magnetometers, able to detect magnetic fields in multiple directions at once, potentially gathering a richer signal if noise challenges can be overcome.
OPMs typically still need magnetic shielding, however. For OPMs these needs are not just due to the extreme difficulty of detecting fields from the brain against the Earth’s much larger field, but also due to tradeoffs in the system’s sensitivity in a noisy ambient field. As a result, many current OPM-MEG systems need almost an order of magnitude more shielding than SQUID systems. With that said, there is substantial potential to overcome these limitations. Due to the sensors’ ability to be placed closer to the scalp, active shielding is a realistic possibility, decreasing the need for passive shielding. By using arrays of OPM sensors to build gradiometers, researchers are able to reduce shielding requirements even more. There are even cases of OPMs being used to measure MEG signals under ambient conditions, albeit with strict restrictions on the environment to make sure that there were no nearby electronic devices or large metal objects that could introduce noise.
OPMs also still lag behind the state of the art in SQUID devices, although they are enjoying rapid progress. Currently, they have lower dynamic range than SQUID MEG, and a higher level of noise. The individual sensors are also bulkier and heated, and there are interference effects between nearby sensors, making it more challenging to make a headset with a large number of them. Still, there has been substantial progress in this area, and at least one system with 384 channels – more than typical SQUID MEG – is in late stages of development.
OPMs can now be routinely used to detect responses to sensory input, and have been used to detect the brain’s tracking of speech rhythms, and even to construct a rudimentary brain-computer interface in which patients spelled words based on looking at letters on a screen.
Next-Generation Magnetometers for MEG 3.0
To get beyond the need for shielding and special-purpose environments entirely, we expect to need a different quantum sensing approach.
Finding the right approach has been a significant journey. We investigated a long list of sensors, including dozens of magnetometers. Some were ruled out quickly, others merited a closer look. In the end, we identified technologies with the potential to go the distance.
Some of the most promising approaches involve solid-state sensors. With dramatically higher density than atomic gases, solid-state systems can pack more sensing power into smaller spaces, while their customizability allows systems to be tailored for a number of desirable properties. Many such systems can work at room temperature, avoiding both the cryogenic requirements of SQUID and the heat of OPMs. Many can also work in large bias fields without the nonlinear effects that degrade OPM performance, allowing them to work in ambient conditions without need for bulky magnetic shielding. These systems can be engineered by implanting defects in a material, as in nitrogen vacancy centers in diamond, by producing artificial crystals like yttrium iron garnet spheres, or by making use of novel properties of layered materials, as in sensors making use of the tunneling magnetoresistance effect or magnetoelectric laminates. While there has been some preliminary progress towards use of these systems for MEG, none of these technologies are currently sensitive enough to reach to the needed femto-Tesla range in a multi-channel system. However, many of these technologies have only recently been explored in this context. We see substantial potential for improvement.
What’s Next?
At e184, we’re building foundational technology for a brain-computer interface for all. Right now, our clearest path forward is to invest in sensor development, and build MEG 3.0.
We’ll be hiring soon. We’ll have physicists and engineers developing sensors, as well as experts in machine learning, building models so we can interpret input from our new sensor technology. We’re also building a board of advisors, looking for experts spanning multiple fields who can keep our project on-track. If you see yourself in those descriptions, send an email to p@e184.com to find out more.