New SMR framework reveals local vs distributed brain signal predictability

A new paper from researchers Maryam Ostadsharif Memar and Nima Dehghani introduces Spatially Masked Regression (SMR), a framework that quantifies how much of a neural recording electrode's signal comes from its immediate neighborhood versus the broader distributed network. Traditional interpretations assume electrodes measure local activity, but signals often carry structured information from across the array. SMR addresses this by reconstructing each electrode's time series from all other electrodes while excluding a configurable mask around the target. By progressively expanding the mask, the method controls spatial locality and reveals how predictive information changes when nearby channels are withheld.

Applying SMR to both intracranial EEG (iEEG) with heterogeneous coverage and scalp EEG over sensorimotor cortex, the authors found strong within-subject reconstruction in both modalities. Even when all local neighbors were excluded, electrodes retained substantial residual predictability, indicating that individual channels reflect both local redundancy and distributed structure. Cross-subject transfer was markedly stronger in EEG than iEEG, suggesting more shared spatial patterns. Surrogate tests that disrupted phase or temporal ordering significantly reduced performance, confirming SMR's reliance on structured temporal and cross-channel organization. The framework provides an interpretable tool for neuroscience, potentially improving brain-computer interfaces and neural network analysis.