Abstract
Recent success in developing optically pumped magnetometers (OPMs) has provided flexible and innovative ways to explore the high spatio-temporal dynamics of the human brain. However, the OPM sensors now in widespread use are highly susceptible to interference magnetic fields in experimental environments. Interference fields not only contaminate the measured neural signals but also introduce channel gain non-uniformity, making it challenging to effectively remove interference signals. Therefore, we propose Variational Bayesian Calibrated Signal Space Projection (VBCSSP), a new signal space projection (SSP) method developed to explicitly incorporate channel gain non-uniformity in its formulation. VBCSSP estimates the channel gain and amplitude of interference signals using hierarchical variational Bayesian estimation. Here, the performance of VBCSSP is comprehensively evaluated using static and dynamic channel gain simulation, a matrix coils (MC) experiment, and a human experimental dataset. The results for the simulation demonstrate that VBCSSP is more robust against channel gain non-uniformity compared to SSP, even when temporal gain fluctuation exists. The results also suggest that the optimal interference model can be selected based on estimated free energy. The properties confirmed in the simulation are also validated in real sensor measurements using the MC data. Finally, in the human experimental dataset, VBCSSP enhances the agreement of the estimated dipoles with SQUID-MEG. VBCSSP is expected to facilitate OPM studies by providing improved interference shielding with channel gain estimation.