Event-related desynchronization (ERD) and synchronization (ERS)

Brain's response to an external or internal stimuli can be manifested in EEG either as classical, phase-locked event-related potentials (ERP), or non-phase locked (induced) reactivity [11]. ERPs are clearly visible on the time average of properly synchronized EEG responses. On the contrary, non-phase locked activity is canceled out from the average. Classically, induced event-related desynchronization and synchronization (ERD and ERS) was quantified by the following procedure: the most reactive frequency bands were chosen by a trial-and-error procedure, and then signals, band-pass filtered within those bands, were squared before averaging (for details see [11]). This procedure not only depends on the arbitrary choice of frequency bands, but gives only an overall time course of energy of the whole band, hiding any subtle effects possibly occurring within it. A complete structure of dynamic evolution of energy, contributing to ERD and ERS, can be calculated by averaging time-frequency distributions of single trials. As discussed above, MP offers here a high-resolution estimate, obtained from a general procedure free of a priori choices, needed e.g. for setting parameters of time-frequency distributions from the Cohen's class (see footnote 4). Detailed comparison of MP-based estimates with the classical approach, and increase of sensitivity especially in the gamma band, is discussed in [4].

Figure 4: Bottom: first 10 of 57 EEG trials (Hjorth source derivation) recorded during a voluntary finger movement. Middle plot--average of the time-frequency distributions of energy obtained from MP parametrization. Horizontal scale in seconds (movement in time 0), vertical--frequency in Hz. Top: the same distribution in 3 dimensions, presentation as in Figure 2. We clearly observe e.g. splitting of the alpha activity into two differently reacting subbbands, a phenomenon elusive to the classical approach.

Figure 4 presents a picture of desynchronization and post-movement resynchronization in alpha and beta bands in a similar experiment concerning voluntary finger movement. Bottom plot contains time courses of 10 of 57 trials used for EEG analysis, finger movement occurring in time "0". Each trial was decomposed by MP with stochastic dictionaries ([3]), and resulting estimates of time-frequency energy density (eq. (7)) were averaged. Resulting picture is presented in two and three dimensions (middle and upper plots). We observe that two distinct alpha components are well resolved, the upper alpha desynchronizes earlier than its lower component. This example highlights advantages offered by the unique time-frequency resolution of the proposed methodology.