Laboratoire de Cartographie fonctionnelle du Cerveau, UNI - ULB Neuroscience Institute, Université libre de Bruxelles (ULB), Brussels, Belgium.
UR2NF - Neuropsychology and Functional Neuroimaging Research Unit at CRCN - Centre de Recherches Cognition et Neurosciences, and UNI - ULB Neuroscience Institute, Université libre de Bruxelles (ULB), Brussels, Belgium.
Neuroimage. 2021 Apr 15;230:117793. doi: 10.1016/j.neuroimage.2021.117793. Epub 2021 Jan 23.
The linearly constrained minimum variance beamformer is frequently used to reconstruct sources underpinning neuromagnetic recordings. When reconstructions must be compared across conditions, it is considered good practice to use a single, "common" beamformer estimated from all the data at once. This is to ensure that differences between conditions are not ascribable to differences in beamformer weights. Here, we investigate the localization accuracy of such a common beamformer. Based on theoretical derivations, we first show that the common beamformer leads to localization errors in source reconstruction. We then turn to simulations in which we attempt to reconstruct a (genuine) source in a first condition, while considering a second condition in which there is an (interfering) source elsewhere in the brain. We estimate maps of mislocalization and assess statistically the difference between "standard" and "common" beamformers. We complement our findings with an application to experimental MEG data. The results show that the common beamformer may yield significant mislocalization. Specifically, the common beamformer may force the genuine source to be reconstructed closer to the interfering source than it really is. As the same applies to the reconstruction of the interfering source, both sources are pulled closer together than they are. This observation was further illustrated in experimental data. Thus, although the common beamformer allows for the comparison of conditions, in some circumstances it introduces localization inaccuracies. We recommend alternative approaches to the general problem of comparing conditions.
线性约束最小方差波束形成器常用于重建神经磁记录所支持的源。当必须跨条件比较重建时，从所有数据一次估计单个“通用”波束形成器被认为是良好的实践。这是为了确保条件之间的差异不是归因于波束形成器权重的差异。在这里，我们研究了这种通用波束形成器的定位准确性。基于理论推导，我们首先表明，通用波束形成器会导致源重建中的定位误差。然后，我们转向模拟，其中我们尝试在第一个条件下重建（真实）源，同时考虑在大脑的其他地方存在（干扰）源的第二个条件。我们估计错位图，并统计评估“标准”和“通用”波束形成器之间的差异。我们用对实验 MEG 数据的应用来补充我们的发现。结果表明，通用波束形成器可能会导致显著的定位错误。具体来说，通用波束形成器可能会迫使真实源被重建得更接近干扰源，而不是真实的位置。由于同样适用于干扰源的重建，两个源都被拉得更近了。这一观察结果在实验数据中得到了进一步的说明。因此，尽管通用波束形成器允许条件比较，但在某些情况下，它会引入定位不准确。我们建议针对比较条件的一般问题的替代方法。
