Time-delayed decorrelation for the identification of cardiac artifact components in MEG data

The time-delayed decorrelation (TDD) Independent Component Analysis (ICA) applied to continuous multichannel magnetoencephalographic (MEG) recordings is a straightforward calculation in the time domain: Its main ingredient is a calculation of the correlation between signals having different time shifts with respect to each other. The equivalent mathematical representation in the frequency domain and a simple simulation show that TDD exploits different complex frequency spectra of the sources. Applying ICA to 80-channel whole head MEG data obtained under visual stimulation we can assign five of the 80 source components to the cardiac artifact. Their complex spectra exhibit a randomly distributed phase showing that TDD is capable of isolating quasi-periodic signals even for the case of an unspecified phase.