Signal space beamforming for gain- and geometry-independent interference suppression in wearable MEG

Optically pumped magnetometers (OPMs) for wearable magnetoencephalography (MEG) offer substantial flexibility compared to cryogenic MEG, but they also introduce new challenges related to environmental noise and movement artefacts. State-of-the-art denoising techniques such as signal space separation require not only dense OPM arrays but also the fine calibration of sensor gain and geometry. This may complicate their use in complex experimental situations involving uncooperative patients or subjects such as newborns or fetuses, where OPM positioning may be unstable or plain unknown. To address this limitation, we introduce signal space beamforming (SSB), a version of beamforming applied to sensor signals and designed to suppress interferences independently of sensor gain and geometry. We show that SSB operates via a trade-off between noise suppression and neural signal preservation that is controlled by a single soft-threshold parameter, which we calibrated using phantom measurements. We then validated the effectiveness of SSB across several datasets. Using interictal OPM-MEG recordings in epileptic patients, SSB successfully cleaned signals while preserving epileptiform discharges. Moving onto cutting-edge early neurodevelopmental OPM-MEG recordings, SSB allowed to recover auditory evoked responses in newborns and fetuses similar to previous findings but with more streamlined preprocessing and higher response amplitude in the case of the fetal data. We contend that SSB provides a pragmatic solution to exploit OPM-MEG data recorded in challenging conditions where geometry-dependent methods cannot be used, opening new frontiers for pioneering clinical and fundamental applications of the OPM technology.