dSPM & noise normalization

Dynamic statistical parametric mapping (dSPM) and related noise-normalized estimators post-process a minimum-norm solution by dividing each source's estimated amplitude by its projected noise standard deviation, computed from a baseline or empty-room noise covariance. The result is a statistic (like a z-score or F-map) at every source, rather than a raw current amplitude.

Noise normalization improves the spatial behaviour of MNE: it partially removes the depth bias, because deep sources with intrinsically low sensitivity also carry larger projected noise, so the ratio is more uniform across depth. dSPM produces smoother, more uniform point-spread across the cortex than raw MNE. It does not, however, add information beyond MNE — it is a rescaling of the same linear estimator — so it cannot resolve sources that the lead field cannot distinguish.