Clinical implications
In the present study, SNRs and (to a lesser degree) amplitudes obtained with QmfrVEPs were correlated with visual field losses in glaucoma patients. This shows that the SNR can be used as a direct outcome parameter to quantify functional losses and not only as an indicator for signal quality.
All segments with SNR values above 10 dB had normal sensitivity in the corresponding visual field locations (“undamaged” segments). However, some normal subjects had SNR values of similar magnitude as in segments that included locations with visual field defects in the glaucoma patients (“damaged” segments). Therefore, QmfrVEP may allow identification of local defects in glaucoma patients and objective demonstration of normal function when SNR values are high, but the prediction of absolute sensitivities in the visual fields may be limited by the large interindividual variability of the VEP.
It has been shown before that reduced mfVEP amplitudes can be an early predictor of axonal damages [
11] and can be used to detect visual field losses in unilateral optic neuritis [
8], temporal hemianopia [
8,
21] and glaucoma [
26]. In addition, patients who display structural damage in OCT images showed changes in mfVEP. Furthermore, the magnitude of mfVEP defects was correlated with retinal nerve fibre layer thickness [
21]. In addition, mfVEP has previously been found to give more comprehensive information about damage in small parts of the visual field, than the VEP [
27].
The small sample size of five patients limits the applicability of parameters like sensitivity and specificity. However, our data show a large effect size of SNR when comparing damaged and undamaged segments.
Currently, mfVEP recordings are not used in clinical routine due to the considerable interindividual variability of the mfVEP amplitudes [
2,
7,
26,
28,
29] and due to the absence of a “gold standard” [
8]. Possibly, the multifrequency technique, as presented in the current study, is a useful alternative to obtain objective field losses because the evaluation of the signals is easier and less prone to errors than measuring single peaks in transient responses. Further studies including larger cohorts of subjects are required to assess sensitivity and specificity for detecting subjects with visual field defects.
The QmfrVEP can possibly be used in other diseases. It would be particularly interesting to study whether the SNR improves in parallel with the visual field in diseases with reversible defects (e.g. compressive neuropathy due to a pituitary tumour measured pre- and postoperatively). Repeated measurements in progressive visual field losses might uncover the diagnostic utility of amplitudes and SNRs in follow-up of these patients.
The response phases can also be obtained from the DFTs but were not analysed in the current study because only phase values measured at the same channel can be compared. Phases strongly vary between different individuals and between different stimulus locations. However, the phases may be an interesting additional parameter in single patients to monitor disease progression or the effects of therapeutic intervention.
Optimizing stimulus parameters
Our measurements show that QmfrVEPs can provide information from several retinal locations simultaneously. Improvement of this technique by an increase of the number of stimulated retinal locations may be a further development to reach an objective perimetry. Earlier, Abdullah and coworkers [
22] were able to demonstrate the possible usefulness of 9 and (to a minor degree) 17 stimulation fields.
In the present study, four visual quadrants were stimulated using checkerboard devices of fixed geometry. Thus, different field sizes of the stimulated retinal areas could only be achieved by changing the distance between stimulus and observer or with stimulators showing other geometric data. If peripheral locations are to be included, the stimuli should correct for eccentricity dependent cortically scaling [
30].
In addition, if more stimulus areas are needed the frequency differences in the stimuli should be small enough to ensure that all VEPs originate in the same cortical mechanism. If a high number of stimulus areas is used, the recording time should be increased accordingly in order to able to achieve sufficient temporal resolution and to distinguish the responses and SNRs elicited by the different arrays. However, it should be noted that, in contrast to the m-sequence-based mfVEP, interruption and restart of measurements are not possible, because a complete measurement is necessary to perform the DFT. Therefore, recordings with large artefacts (especially where the signals were outside of the amplifier’s limits) must be considered carefully. An improvement of the SNR can be achieved by averaging repeated traces before spectrum analysis.
In addition, all stimulus frequencies should be sufficiently different from those at which alpha waves occur (i.e. at about 10 Hz) [
12] and should be chosen outside this frequency range. Preceding studies showed that a stimulus frequency of about 12 Hz, as used in the present student, gives optimal results [
19,
28,
31‐
33].
As in an earlier study [
19], we used the ratio of signal amplitude at reversal frequency divided by the mean of two neighbouring frequencies for calculations of the SNR ratio [
14]. Pilot studies showed that the use of two direct neighbouring frequencies for SNR calculation performed better than the inclusion of 4 or more [
34] non-signal amplitudes.
One disadvantage of the method is on the one hand the long duration (50 s per run) and on the other hand the high number of repeats. If we succeed in using dichoptic stimuli, it will allow us to reduce the test duration. A disadvantage of the objective technology in comparison to standard perimetry is the time-consuming preparations for placing the electrodes. Compared to the usually applied electrode configuration in multifocal VEP measurements, we used one additional central electrode. This additional electrode was most frequently (27.5%; see Fig.
3) chosen as best channel from the best-of-algorithm. However, if this additional electrode was excluded from correlation between SNR and perimetric losses only a small reduction of the correlation coefficient was seen (seven channels: − 0.76, six channels: − 0.73).
Finally, it should be taken in consideration that our LED arrays used light tight material between the LEDs achieving a contrast of nearly 100% [
19]. This could mask visual defects of patients with glaucoma disease, because of saturating damaged and undamaged quadrants simultaneous. Therefore, future measurements with reduced stimulation contrasts [
22] might be helpful to improve discrimination.