Optimal configuration of on-scalp OPMs with fixed channel counts

Recent technological developments have brought optically pumped magnetometers (OPMs) within reach of the larger neuroscientific community. The current state-of-the-art consists of whole-head systems that measure the magnetic field at >100 locations. OPM sensors can be constructed to measure the field in either 1, 2, or 3 orientations. Consequently, the number of channels can differ from the number of sensors. This allows for magnetoencephalography (MEG) system designs with multiple measurement orientations at fewer locations, many locations with fewer orientations, or, ideally, many locations with multiple orientations. Yet, due to budget constraints, starting OPM groups are typically getting fewer sensors than what could, in principle, be accommodated in a whole head helmet-like arrangement. Furthermore, implementing multiple orientations in a single sensor comes at a cost and hardware companies are still optimizing the trade-offs between sensor designs. To guide the OPM systems design, it is relevant to know the optimal spatial distribution and sensing orientation of OPMs. We performed a simulation study in which we kept the total number of channels constant. We compared 3 synthetic 192-channel OPM arrays that were composed of either monoaxial, biaxial or triaxial sensors, where the sensors were placed at either 192, 96, or 64 measurement locations, respectively. We simulated multiple instances of an MEG signal due to a dipolar source in the brain, contaminated by various combinations of noise, considering sensor noise, brain noise, and noise induced by head (and sensor) movements in the residual ambient magnetic field. An optimal design of the MEG system serves both to record the activity of the brain, as well as the environmental noise that is to be suppressed. We performed dipole fits and evaluated the localization error and the amplitude of the estimated dipole moment. We cleaned the data using various spatio(temporal) cleaning strategies prior to fitting the dipoles. Our observations confirm earlier work, in that 1) the sensing orientation radial to the head is in general more optimal to pick up activity from the brain than tangential directions, but that 2) adding sensing orientations tangential to the head surface helps in suppressing ambient noise sources. Yet, we did not observe a clear improvement comparing triaxial with biaxial OPMs. Given that triaxial sensing may come at the expense of reduced spatial sampling over the head and reduced signal-to-noise for individual channels, we conclude that, given a fixed number of channels, biaxial sensors may be preferred with the currently available technology.