Source Leakage and Leakage Correction

Any linear inverse or beamformer maps sensor data to source estimates through a point-spread function, so the estimated time course at one source location is contaminated by leakage from its neighbours. Because that leakage is instantaneous, it manufactures strong spurious zero-lag correlation between nearby source estimates, which would be mistaken for functional connectivity if computed naively. Leakage is the single biggest reason source-space connectivity is treated with suspicion.

Two strategies are standard. The first uses connectivity metrics that are by construction insensitive to zero-lag coupling (imaginary coherency, PLI, wPLI), accepting their blindness to true instantaneous interactions. The second explicitly removes shared zero-lag signal before estimating connectivity, for example by symmetric orthogonalization of the set of source time courses (Colclough), which pairwise or jointly regresses out the instantaneous shared component. Both approaches necessarily discard any genuine zero-lag interaction along with the artifact, which is the accepted price of trustworthy source connectivity.