Yttrium-Iron Garnet Magnetometer in MEG: Advance towards Multi-Channel Arrays

Ekaterina Skidchenko, Anna Butorina, Maxim Ostras, Petr Vetoshko, Alexey Kuzmichev, Nikolay B. Yavich, Mikhail S. Malovichko, Nikolay A. KoshevSensors 2023 · OpenReview

TL;DR

This paper investigates the potential of a new type of sensor, the Yttrium-Iron Garnet Magnetometer (YIGM), for magnetoencephalography (MEG). While YIGMs show promise due to their proximity to the scalp, their current noise levels are too high for practical multi-channel MEG systems, highlighting the need for further noise reduction.

What Problem Does This Paper Address?

Magnetoencephalography (MEG) is a powerful neuroimaging technique that measures the faint magnetic fields produced by brain activity. However, these fields are incredibly weak (fT-order) and decay rapidly with distance. This necessitates highly sensitive sensors placed very close to the scalp. Traditional MEG systems rely on Superconducting Quantum Interference Devices (SQUIDs), which require bulky cryogenic cooling and are limited in their placement flexibility. Optically Pumped Magnetometers (OPMs) offer an improvement by operating at room temperature and allowing for more flexible arrays, but they still have limitations like a narrow dynamic range and heating issues. This paper explores a new contender, the Yttrium-Iron Garnet Magnetometer (YIGM), which is a solid-state, room-temperature sensor with a promising sensitivity. The core problem addressed is evaluating the feasibility and potential of YIGM-based multi-channel MEG systems by analyzing their information-theoretic performance compared to existing technologies.

Methods

The researchers employed a simulation-based approach using lead-field matrices. They modeled different multi-channel sensor layouts for YIGMs, as well as for SQUID and OPM systems. The core of their analysis involved calculating key performance metrics:

  • Signal-to-Noise Ratio (SNR): This measures how well the sensor can distinguish a true signal from background noise.
  • Total Information Capacity (TiC): Derived from information theory, this metric quantifies the overall amount of information an array of sensors can capture about brain activity.

They used realistic noise levels for each sensor type and considered various sensor placements (normal and tangential to the scalp) and array configurations (SQUID-like and uniform distributions). The analysis focused on comparing the relative SNR and TiC of YIGM systems against established SQUID (Elekta Neuromag) and OPM systems.

Datasets

The study utilized a single high-resolution structural T1-weighted MRI scan for creating an anatomical model of the head. This model was used to generate the lead-field matrices, which are crucial for simulating the magnetic field propagation from brain sources to sensors. No new datasets were introduced or benchmarked in this paper.

Key Insights and Ideas

  • Proximity is Paramount: YIGMs' ability to be placed extremely close to the scalp (down to millimeters) gives them a significant advantage in SNR and TiC compared to SQUIDs and OPMs, even at their current noise levels.
  • Noise is the Bottleneck: Despite the proximity advantage, the current intrinsic noise level of YIGMs (around 35 fT/√Hz) is too high to create a truly competitive multi-channel MEG system. Significant noise reduction is the primary requirement for YIGM-based MEG to surpass existing technologies.
  • Gradiometers Show Promise: YIGM-based gradiometer configurations appear to offer better performance than magnetometers, aligning with established MEG principles where gradiometers are often preferred for cortical source localization.
  • Tangential vs. Normal Orientation: While tangential YIGM placements offer closer proximity, they surprisingly yielded poorer results. This is attributed to the magnetic field's rapid decay and how volumetric currents can suppress the signal amplitude in tangential measurements.
  • Channel Count Matters, But Noise Dominates: Increasing the number of sensors in an array generally improves coverage and performance, but this effect is less pronounced than the impact of reducing sensor noise.

Key Results

  • At their current noise levels, YIGMs outperform SQUIDs and OPMs in terms of SNR and TiC due to their closer proximity to the scalp.
  • To outperform the conventional Elekta Neuromag SQUID system with 102 magnetometers, YIGM noise levels need to be reduced to below 5 fT/√Hz for SNR and below 13 fT/√Hz for information capacity.
  • For gradiometers, a YIGG layout with a noise level of 10 fT/√Hz can outperform SQUID gradiometers. With further noise reduction to 3 fT/√Hz, YIGG systems could significantly outperform SQUID gradiometers.
  • Even with 333 sensors, YIGMs require a noise level reduction to one-third of their current value to match the information capacity of the Elekta Neuromag system.

Limitations and Future Work

  • Current Noise Levels: The most significant limitation is the current noise level of YIGM sensors, which is insufficient for constructing a high-performance multi-channel MEG system.
  • Cross-talk: The study assumes no cross-talk between YIGM sensors, a simplification that may not hold true in a densely packed array. Future work should investigate this phenomenon.
  • Tangential Layouts: The poorer performance of tangential YIGM layouts warrants further research to fully understand the underlying physics.
  • YIGM vs. OPM Coverage: While YIGM vs. SQUID maps showed excellent coverage, YIGM vs. OPM coverage was less distinct, suggesting a need for further investigation into this comparison.

The authors emphasize that future work should focus on reducing YIGM sensor noise. They also plan to explore more complex sensor placement strategies, investigate averaged head models, and delve deeper into the effectiveness of tangential YIGM layouts. The potential for YIG-MEG systems to outperform current technologies by a significant margin, especially when approaching the theoretical noise limit of 1 fT/√Hz, serves as strong motivation for continued development.

Summary updated

Abstract

Recently, a new kind of sensor applicable in magnetoencephalography (MEG) has been presented: a solid-state yttrium-iron garnet magnetometer (YIGM). The feasibility of yttrium-iron garnet magnetometers (YIGMs) was demonstrated in an alpha-rhythm registration experiment. In this paper, we propose the analysis of lead-field matrices for different possible multi-channel on-scalp sensor layouts using YIGMs with respect to information theory. Real noise levels of the new sensor were used to compute signal-to-noise ratio (SNR) and total information capacity (TiC), and compared with corresponding metrics that can be obtained with well-established MEG systems based on superconducting quantum interference devices (SQUIDs) and optically pumped magnetometers (OPMs). The results showed that due to YIGMs’ proximity to the subject’s scalp, they outperform SQUIDs and OPMs at their respective noise levels in terms of SNR and TiC. However, the current noise levels of YIGM sensors are unfortunately insufficient for constructing a multichannel YIG-MEG system. This simulation study provides insight into the direction for further development of YIGM sensors to create a multi-channel MEG system, namely, by decreasing the noise levels of sensors.
BibTeX
@inproceedings{skidchenko2023yttrium,
  author = {Ekaterina Skidchenko and Anna Butorina and Maxim Ostras and Petr Vetoshko and Alexey Kuzmichev and Nikolay B. Yavich and Mikhail S. Malovichko and Nikolay A. Koshev},
  title = {{Yttrium-Iron Garnet Magnetometer in MEG: Advance towards Multi-Channel Arrays}},
  booktitle = {Sensors},
  year = {2023},
  url = {https://openreview.net/forum?id=Qz9MQufipr},
}
Figure