Parametric Data-driven Statistical Test for Detection of Instantaneous Functional Coupling from MEG Data

Noninvasive assessment of functional connectivity is prevented by the artifacts of volume conduction. Traditionally, in order to overcome this problem, the methods based on the exclusion of the real part of the cross-spectrum are used. Together with instantaneous artifacts of volume conduction, these methods exclude true zero-phase coupling as well. Since zero and near-zero phase lag coupled activity is present in the brain and can be conditioned by bidirectional communication, dynamical relaying or common input, other approaches that preserve this type of coupling are in great demand. One of these approaches is implemented with the use of PSIICOS-projection. This projection can be applied to the vectorized sensor-space cross-spectrum. The debiased cross-spectrum is projected to the source space by the means of inverse mapping. This results in the estimates of the pairwise source-space cross-spectral coefficients. The resulting coefficients cannot be normalized, since PSIICOS-projection distorts the power component. Therefore, the coefficients are influenced by the power of the corresponding sources. To cope with this we propose a novel statistical testing framework. Within this study we developed and compared several versions of statistical tests. They implied that the null distribution of cross-spectral coefficients is modeled with Wishart distribution. The most optimal version of the test was based on sampling in source space with the scale matrix, the off-diagonal terms of which were equal to zero. Realistic simulations demonstrated high performance of the developed approach. The additional set of simulations demonstrated that the developed test excludes the pairs characterized by the powerful but not truly coupled sources. Applied to the real data from mental rotation task, the method revealed functionally relevant networks and excluded the implausible networks. Thus, the proposed framework can be reliably used for estimation of low phase-lag coupled brain sources.