Nonlinear prediction as a tool for tracking the dynamics of single-trial readiness potentials

The performance of a voluntary act is preceded by an intrinsic process of intention and preparation accompanied by specific patterns of scalp-recorded EEG. The baseline shift to negativity and the decrease of alpha and beta oscillations prior to movement performance are considered to reflect the motor preparation and are observable even in single-trial EEG records during repetitive voluntary movements. The single-trial features of these patterns are of importance since they would reflect the dynamics of EEG activity during movement performance throughout the experimental session. Since this process is very complex and its dynamics is unclear, we applied a more general approach, i.e. the method of nonlinear prediction (NP) recommended for detection of chaos in the behavior of nonlinear dynamical systems. The NP is based on a library of past patterns of the time series used to compute the prediction of future pattern. If the time series is chaotic, the correlation coefficient between predicted and actual values would decrease as a function of the prediction step. The NP was applied to EEG records containing readiness potentials (RPs) at Fz, Cz, Pz, C3 and C4 during repetitive voluntary finger flexion. Because of nonstationarity, the prediction and the library spanned the same time period. As a result (1) chaotic segments in EEG records were detected; and (2) the sharp jumps of the absolute error between predicted and actual values indicated instances of poor prediction, which were interpreted as indicating new stages of EEG activity related to the preparatory process of motor activity.