Data-driven model optimization for optically pumped magnetometer sensor arrays

Area of Science:

• Biophysics
• Neuroimaging
• Sensor Technology

Background:

• Optically pumped magnetometers (OPMs) are emerging as sensitive, portable alternatives to traditional superconducting systems for magnetoencephalography (MEG).
• OPMs eliminate the need for cryogenic cooling, enabling direct scalp placement.
• The distinct physical principles of OPMs introduce novel modeling challenges for sensor array design and data analysis compared to established superconducting quantum interference device (SQUID) systems.

Purpose of the Study:

• To develop and validate an empirical Bayesian framework for comparing and optimizing optically pumped magnetometer (OPM) sensor arrays for magnetoencephalography (MEG).
• To investigate the feasibility of estimating OPM sensor geometry directly from recorded brain data, thereby potentially relaxing the requirement for precise a priori sensor localization.

Main Methods:

• Simulated perturbation of sensor geometry to estimate true sensor configuration using analytic model comparison.
• Utilizing width of perturbation curves for comparing different MEG system configurations.
• Testing the framework with simulated and real MEG data from both SQUID and OPM recordings, using head-casts and scanner-casts.

Main Results:

• The proposed Bayesian framework effectively compares and optimizes OPM sensor arrays.
• The technique successfully estimates true sensor geometry from simulated and real OPM data.
• Demonstrated that accurate sensor geometry can be inferred from OPM data using a model comparison framework, leveraging cortical manifold information.

Conclusions:

• The developed empirical Bayesian framework provides a robust method for optimizing OPM sensor arrays in MEG.
• It is possible to estimate OPM sensor geometry directly from MEG data, reducing reliance on pre-acquisition localization.
• This approach simplifies MEG system setup by potentially removing the need for precise a priori sensor positioning and co-registration procedures.