Amplitudes and latencies of single-trial ERP's estimated by a maximum-likelihood method.

The common approach in research on event-related electroencephalogram (EEG) potentials is to assume that the trigger-related signal is always the same and can be extracted from EEG background activity by simple averaging. To check the validity of this approach and to provide more exact results, latencies and amplitudes of components have to be estimated in single trials. Pham et al. applied a maximum-likelihood approach to solve the more general model which assumes that the signal hidden in EEG background activity has the same shape and amplitude but may vary in its latency from trial to trial. Extending their method we present a solution in which amplitude variability is also allowed. The utility of the solution to estimate the P3 component in single trials was investigated both by extensive pseudoreal simulations and in an application to real data. The simulations showed some advantage of the method over two other methods (Woody's method and peak-picking) commonly used in event-related potentials research. Application to real data provided a plausible description of single-trial sequential effects on the amplitude of the P3 component.