Array-gain constraint minimum-norm spatial filter with recursively updated gram matrix for bioelectromagnetic source imaging

Abstract

Event Abstract Back to Event Array-gain constraint minimum-norm spatial filter with recursively updated gram matrix for bioelectromagnetic source imaging Kensuke Sekihara1* and Isamu Kumihashi1 1 Tokyo Metropolitan University, Department of Systems Design and Engineering, Japan The spatial filter is a popular method used to reconstruct a source distribution from bioelectromagnetic data. When the weight of a spatial filter only depends on the geometry of the measurements, such a spatial filter is referred to as a non-adaptive spatial filter, which includes the well-known minimum-norm filter [1]. The weight of an adaptive spatial filter depends not only on the measurement geometry but also on the measurement covariance matrix. A representative adaptive spatial filter is the minimum-variance spatial filter [2]. Although adaptive spatial filters generally have performances better than non-adaptive spatial filters, adaptive spatial filters have well-known weaknesses. First, since the computation of their weights requires a sample covariance matrix of the measured data, a large number of time samples are needed to obtain an accurate sample covariance matrix. Second, the adaptive spatial filter is known to be sensitive to the source correlation, and it generally fails to reconstruct source activities when they are highly correlated. In this paper, we propose a novel spatial filter based on the array-gain constrained minimum-norm filter, which is a modified version of the minimum-norm filter. The proposed method is designed to estimate the theoretical form of the measurement covariance matrix through recursively updating the gram matrix. The method is able to provide three-dimensional volume reconstruction with a spatial resolution considerably higher than that of existing minimum-norm-based methods. Nonetheless, the method is free of the above-mentioned weaknesses occurring with adaptive spatial filters, because it does not use the sample covariance matrix. The method is also robust to the source correlation, and this robustness can be seen in our computer simulation as well as in our experiments using auditory-evoked MEG data.