We recently released a preprint titled ‘Magnetomyography: A novel modality for non-invasive muscle sensing.’ We wrote this post as a guide to the contents of the paper, detailing why we did this work, what we learned, and its relevance to next-generation sensing for human-computer interfaces.
Magnetomyography, or MMG, is a technique for measuring muscle activity through the detection of magnetic fields. It’s directly relevant to the first product we’re developing at Sonera, the S1, which we’re building to make MMG widely available. Since MMG isn’t well-studied (there are only a couple dozen published papers), we felt it was important to do a deeper dive into this modality alongside development of the S1. The work we’re sharing today marks the beginning of this effort, focusing on basic characterization of MMG and how it compares to its far more popular electrical counterpart, surface electromyography, or sEMG.
In this work, we used third party magnetic sensors in ideal conditions and had subjects perform a variety of motor tasks while we measured both MMG and sEMG signals. Here are the key takeaways from our work:
1. MMG is similar enough to sEMG to be an alternative approach for detecting muscle activity. Both have similar characteristics in terms of spectral content (i.e. the data from both modalities looks the same when overlaid) and both are direct measures of muscle contraction (i.e. there’s a strong correlation between signal power and measured force). That means MMG can be used for applications currently served by sEMG to detect varying amounts of muscle contraction, such as gesture control, prosthetics and rehabilitation devices.
2. MMG is distinctly different from sEMG in ways that can be leveraged to unlock novel capabilities. While comparable to sEMG, MMG is its own unique dataset that contains information not captured in sEMG. For example, MMG signals show higher power at higher frequencies (>100Hz) and are more sensitive to sensor placement relative to the source of the signal. These differences indicate potential for higher specificity and better localization. Possible use cases served by these differences are those requiring high frequency content and high spatial specificity, such as detection of muscle fatigue or assessment of motor unit action potentials.
3. MMG-based systems require new design considerations. MMG can operate at the same fidelity as sEMG without direct skin contact, making this modality more robust to real-world conditions. However, magnetic fields decrease with distance from the source, something that has to be considered when designing MMG-based systems. Our findings show that MMG signals decrease slower than we expected based on prior literature, allowing us to measure MMG signals at distances as far as 50 mm away from the skin. So while it’s true that the signal strength drops off with distance, there’s a high tolerance for barriers (i.e. a watch strap) between the source of the signal and the sensor and an opportunity to detect deeper muscle activity.
4. Sensor technology is a major limiting factor when it comes to scaling MMG. The sensors we used in this study have various limitations. Some are just at the threshold of being able to pick up MMG signals; some are unable to capture MMG signals at low contraction strengths; some are unable to capture the full frequency range of interest due to bandwidth limitations; and some need extensive shielding to operate (and we’re not even talking about size or cost here, which is relatively high for all of them). That means the key to scaling MMG lies in building a sensor that can fully capture MMG signals in ambient conditions, while also being cheap and easy to mass produce - hence the motivation for building the S1.
The findings above reveal the power of a novel sensing modality like MMG and indicate how it could address some of the limitations of sensor modalities like sEMG.
What we’re interested in next is understanding how this impacts the development of neural control systems, like gesture-based control for personal computing. If we can suddenly measure higher quality signals in a more robust way, how will that translate into a better experience for the user? And if we can create a better experience, how will this shift the landscape of personal computing and the types of applications that can be built?
We’ll be sharing more work this year as we continue to answer these questions (and many others) for ourselves, so stay tuned :)
--
Nishita Deka, CEO
Richy Yun, Lead Neuroscientist
