Single trial ERP estimation in the frequency domain using orthogonal polynomial trend analysis (OPTA): estimation of...

A new Event Related Potential (ERP) model is described in which ERPs are estimated as single cortical responses. This model is based on an Orthogonal Polynomial Trend Analysis of Variance (OPTA) in the frequency domain. The orthogonal polynomials are limited to the fifth order in this study. The means and trend-coefficients which are used to estimate the ERPs are tested for significance with the F-statistics. Coefficients not reaching a preset significance criterion are set to zero. Subsequently significant means and trend-coefficients multiplied with the polynomials are inversely transformed to the time domain for each trial. The method is demonstrated with artificial signal-in-noise EEG (electroencephalogram) series in which the signal component (the ERP) decreases in a relatively short time span. The estimated signals approximate closely the input signals and a signal-to-noise improvement on single trial level is reached of about a factor 10. Comparison with the statistical 'Wiener' filtering method shows that the OPTA detects weaker signals. Single trial ERPs estimated with the OPTA from EEGs collected in a habituation paradigm suggest that the averaged evoked potential model may be inadequate in detecting ERP change over a few trials. It is shown that the frontal P3b wave habituated in 7-9 trials, a habituation speed which is comparable to the habituation rate of the visual orienting reaction and the electrodermal orienting reaction.