Imaging of neural oscillations with embedded inferential and group prevalence statistics
Fig 2
a) Examples of designs: The experimental design (shown as a black trace) determines the quality function f(s), so that this latter takes high values for signals consistent with the hypothesis (in orange; the signals that do not correspond to the tested hypothesis are shown in blue). b) MEG data: the multichannel MEG recordings are captured in the matrix X = {x[t = 1], …, x[t = T]}. c) Computing the signal subspace: spatial patterns P = {p1, …, pD} are extracted from the MEG data by optimizing the quality function with respect to spatial filters W = {w1, …, wD}. Whereas W is used to extract the signals of interest from the multichannel MEG data, P are the forward fields of these signals as they contribute to the measured MEG data. d) Computing the forward model: shown are the MEG spatial patterns G(ρ) generated by two tangential dipoles at location ρ in a single subject. e) Subspace correlation as a scanning metric: The spatial patterns from c) and d) span a subspace of the MEG sensor space. A grid of source locations is scanned with a subspace correlation metric [6], quantifying the smallest possible angle between the data and source subspaces. This yields a distributed map of scores, which highlights possible source locations consistent with the hypothesis.