A signal-to-noise analysis of multifocal VEP responses: an objective definition for poor records

Sixty local VEP records, called the multifocal VEP (mfVEP), can be obtained over a wide retinal area. From subject-to-subject, from day-to-day, and from location-to-location, these records can vary in quality presenting a challenge to quantitative analyses. Here three procedures are described for specifying the quality of mfVEP recordings in terms of signal-to-noise ratios. Monocular mfVEPs were recorded in two, 7-min runs. A '2-run signal-to-noise ratio' (2rSNR) was obtained as [RMS(RunA+RunB)]/[RMS(RunA-RunB)]-1, where RMS is the root-mean-square amplitude of the response over the period from 45 to 150 ms (signal window). Two 'noise-window signal-to-noise ratios' were obtained with the same numerator as the 2rSNR but with the denominators based upon the RMS of a signal-free window from 325 to 430 ms. In one case, inSNR, the denominator was the RMS of the record's noise window and in the other case, mnSNR, the denominator was the mean of the RMS amplitudes of all the signal-free noise windows for the subject. The SNRs were related to false-positive rates (i.e., detecting a signal when none was present) by recording mfVEPs with some of the sectors of the display occluded. In particular, the outer three rings (36 sectors) of the display were occluded so that only noise was recorded; false-positive rates for different values of SNR were calculated. The 2rSNR had the highest false-positive rate largely due to alpha in the records of some subjects. The mnSNR had a lower false-positive rate than did the inSNR because there was little correlation between the RMS of the noise in the signal-free window and the RMS of the noise within the signal window. Use of the mnSNR is recommended over the 2rSNR, especially where alpha contamination can not be eliminated. Ways to improve the SNR of the records are discussed.