Denoising improves visualization of evoked potentials with limited number of trials

A simple metric to study the mechanisms generating event-related potentials

BACKGROUND

There is an active debate about the mechanism underlying the generation of event-related potentials, and, particularly, whether these are generated by additive components, independent of the background EEG, or the phase-resetting of ongoing oscillations.

METHOD

We present a new metric to evaluate trial-by-trial covariations of successive ERP components. Our main assumption is that if two successive ERP components are generated by phase-resetting of a unitary oscillation, they should be time-locked to each other and their single-trial latencies should covary. In contrast, if the components are generated by independent additive components, single-trial latency covariations should not be observed. To quantify the covariance between the single-trial latencies, we define a metric based on latency-corrected averages, which we applied to both simulated and real ERPs.

RESULTS

For the simulated data, there was a clear distinction in latency covariation between the ERPs generated with unitary phase-resetting versus additive models. For real visual and auditory ERPs, we observed a lack of latency covariation of successive components.

COMPARISON WITH EXISTING METHODS

The new metric is complementary to other approaches to study the mechanisms underlying ERP generation, and does not suffer from potential caveats due to filtering artifacts. Moreover, the method proved to be more sensitive than another estimation of single-trial latency covariations using the cross-correlation function.

CONCLUSION

The observed lack of latency covariation shows the presence of parallel, independent processing within each cortical sensory pathway.