Magnetoencephalography (MEG) provides a promising, non-invasive medium for characterizing high temporal brain dynamics and functional connectivity (FC) networks. MEG based studies often include a source space signal reconstruction from the sensor data. However, spatial filters or inverse solutions that are used for source signal reconstruction induce a linear interaction between the source signals, and this phenomenon is often termed as a signal leakage. Therefore, the observed FC between the source signals may be partially or entirely due to signal leakage. In the present study, we demonstrated the usefulness of independent component analysis (ICA) to orthogonalize reconstructed source signals to suppress signal leakage before FC computation. We also compared the ICA based orthogonalization approach to commonly used linear regression, and orthogonalization in complex domain (Hipp orthogonalization) approaches.

We used band-limited (i.e. alpha and beta frequency bands) amplitude time series correlation as a measure of FC. Before FC computation, the seed and test voxel time series were orthogonalized using ICA, linear regression and Hipp orthogonalization methods to suppress the signal leakage. We performed an evaluation using simulated MEG data and real resting state MEG data.

Analysis of simulated MEG data indicates that ICA based orthogonalization approach effectively suppressed the spurious FC due to signal leakage where the false positive rate was as per expectation. Analysis of real resting MEG signal showed that ICA based orthogonalization efficiently suppress the spurious FC for test voxels located nearby to the seed voxel where the signal leakage is very much likely. However, it didn't suppress the actual FC for test voxels located remotely from the seed voxel where the signal leakage is unlikely.

ICA based orthogonalization offers a straightforward and efficient alternative tool to alleviate the spurious FC due to signal leakage in reconstructed MEG source signals.