Using correlation analysis to assess the reliability of evoked potential components identified by signal averaging

Background: Signal averaging is the conventional method of enhancing the signal-to-noise ratio in recordings of evoked potentials (EPs) from the skin surface in humans. However there may be difficulties in reliably identifying the features of interest especially when a low number of trials are averaged or when large artefacts with similar waveform as the signal are present. New Method: This method uses the median cross-correlation coefficient of every possible pair of recording repetitions within user-defined windows across the time course of the evoked potential. This is shown to be a good surrogate of signal-to-noise ratio and can be superimposed on the averaged trace to provide reliability information for all components of interest. Results: We applied our method both to simulated signals embedded within the noise component of EP recordings and to real examples of somatosensory EPs. We demonstrated that it can assess the reliability of recorded components independently from their amplitude and could identify artefacts which mimicked genuine components. Comparison with existing methods: There have been a number of previous approaches to this problem but none has found widespread support. This method adds additional information to a common existing technique and is easy to interpret and apply. Conclusions: This method is used as a visual adjunct to the existing interpretation of averaged evoked potentials and will allow judgements of the reliability of each observed component to be made. This is particularly valuable for situations where few repetitions are possible such as nociceptive evoked potentials, which are of increasing clinical interest.