Detection of Evoked Responses by Bootstrap Methods: New Parameters and ROC Analysis

The aim of this work is to describe and evaluate extensions to a bootstrap technique for detecting ABRs. These methods give significance levels (p-values) for: 1) the correlation coefficient between repeated measurements of the ABR, and 2) the mean absolute value of the ABR. Monte-Carlo Carlo simulations of the EEG with and without added `responses' were used to estimate ROC curves and compare the techniques. The methods were also used to estimate hearing thresholds in 12 normal-hearing subjects. The correlation coefficient was found to provide rather poorer results than the mean absolute value both in simulations and in human recordings. The latter method gave hearing thresholds that are similar (or better) than human observers in visual inspection of the ABR.