On-line statistical detection of average evoked potentials: application to evoked response audiometry (ERA).
An objective method of EP detection in averaged EEG epochs is described which is based on the statistical properties of the averaged EEG in the absence of time-locked EP activity. The statistical properties of the subject's averaged background EEG are summarized by making a set of comparisons between two consecutive EEG epochs immediately preceding each stimulus presentation. After every stimulus presentation, a second set of comparisons is made between the post-stimulus EEG epoch and the immediately preceding pre-stimulus EEG epoch. These two sets of comparisons are then examined to determine whether the latter differ significantly from the former. The technique has been programmed for on-line use on a 12-bit minicomputer, validated on cooperative adult subjects and used for ERA threshold determinations in young children. Using response detection criteria yielding an empirically determined false positive error rate of between 1 and 4%, threshold estimates averaged about 10 dB higher than psychophysical thresholds. The efficiency with which such ERA thresholds may be obtained with this method is substantially greater than that of conventional ERA procedures employing subjective evaluation of averages. Further, the technique allows variation of detection criteria to suit non-audiometric needs and empirical determination of the false-positive error rate under any set of conditions.