Optimizing OPM-MEG Sensor Layouts Using the Sequential Selection Algorithm with Simulated Sources and Individual Anatomy

TL;DR

This paper introduces a new method to optimize sensor placement in OPM-MEG systems using individual anatomy and simulations, improving signal capture and reducing sensor count.

Contribution

A simulation-driven sequential selection algorithm for OPM-MEG sensor layout optimization using individual anatomical data.

Findings

Optimized sensor layouts capture most full-head MFM information with only 15-20 sensors (CC > 0.95).

Localization errors for auditory responses were less than 5 mm using equivalent current dipoles.

SSA performance is robust to individualized head geometries, supporting clinical applications.

Abstract

Magnetoencephalography (MEG) based on optically pumped magnetometers (OPMs) offers the flexibility to position sensors closer to the scalp, which improves the signal-to-noise ratio compared to conventional superconducting quantum interference device (SQUID) systems. However, the spatial resolution of OPM-MEG critically depends on sensor placement, especially when the number of sensors is limited. In this study, we present a methodology for optimizing OPM-MEG sensor layouts using each subject’s anatomical information derived from individual magnetic resonance imaging (MRI). We generated realistic forward models from reconstructed head surfaces and simulated magnetic fields produced by equivalent current dipoles (ECDs). We compared multiple simulation strategies, including ECDs randomly distributed across the cortical surface and ECDs constrained to regions of interest. For each simulated magnetic field map (MFM) database, we applied the sequential selection algorithm (SSA) to identify sensor positions that maximized information capture. Unlike previous approaches relying on large measurement databases, this simulation-driven strategy eliminates the need for extensive pre-existing recordings. We benchmarked the performance of the personalized layouts using OPM-MEG datasets of auditory evoked fields (AEFs) derived from real whole-head SQUID-MEG measurements. Our results show that simulation-based SSA optimization improves the coverage of cortical regions of interest, reduces the number of sensors required for accurate source reconstruction, and yields sensor configurations that perform comparably to layouts optimized using measured data. To evaluate the quality of estimated MFMs, we applied metrics such as the correlation coefficient (CC), root-mean-square error, and relative error. Our results show that the first 15 to 20 optimally selected sensors (CC > 0.95) capture most of the information contained in full-head MFMs. Additionally, we performed source localization for the highest auditory response (M100) by fitting equivalent current dipoles and found that localization errors were < 5 mm. The results further indicate that SSA performance is insensitive to individualized head geometry, supporting the feasibility of using representative anatomical models and highlighting the potential of this approach for clinical OPM-MEG applications.

Figures