Sparse Sensor Layout Design via Recursive Orthogonalization of the Forward Solution Matrix With a Realistic Noises Environment in MEG

The trend toward sensor miniaturization has heightened interest in optimizing sensor array configurations across various scientific and industrial applications that imply multichannel measurements. In this article, we present a novel method for sensor array layout optimization for the needs of the modern real-life magnetoencephalographic (MEG) applications. Starting from a superset of potential locations, we form a multisensor probe comprising sensors placed at the given number of distinct locations so that the signal-to-noise ratio (SNR) of the resulting multichannel array is maximized. We achieve this using a fixed number of iterations equal to the number of available sensors. At each step, the method places a sensor to the location that maximizes the region of interest (ROI)-related SNR and then applies the projection operation to the rows of the forward model matrix to orthogonalize the subsequent stereotypic iterations with respect to the sources served by the already-selected sensors. Within a selected ROI, the developed approach allows for the performance comparable to that of 102-sensor industrial standard (Elekta Neuromag MEG system) in terms of SNR that is from −0.83 to 2.13 dB for different types of compact sensors. Our approach requires significantly less computational resources and is $50\times $ – $70\times $ faster as compared to the previously developed methods. Due to high flexibility, RALFE sparse sensor design, demonstrated in application to MEG, is readily applicable to many other multichannel measurement challenges with linear observation models.