Benchmarking of Averaging Methods Using Realistic Simulation of Evoked Potentials

Abstract

The objective of this research is to conduct a comparative evaluation of various averaging methods for estimating evoked potentials using realistic simulations. Simulated signals are commonly employed to assess pattern recognition algorithms for event-related potential estimation since obtaining gold standard records is challenging. The simulations used are considered realistic because they allow for variations in potential latency, component width, and amplitudes. Background noise is simulated using an 8th order Burg autoregressive model derived from the analysis of a real dataset of auditory evoked potentials. The simulations incorporate actual instrumentation and acquisition channel effects, as well as power line interference. Three averaging methods for estimating the evoked potential waveform are compared: classical consistent average, weighted average, and reported average. The comparisons are conducted in two scenarios: one without artifacts and the other with 20% contamination by artifacts. The results of the comparative evaluation indicate that the trimmed average offers the best trade-off between the estimated signal-to-noise ratio (SNR) value and bias.