Objective Visual Acuity Estimates in Amblyopia Are More Accurate With Optotype-Based P300 Than With VEP Measurements

Abstract

Purpose: Traditional visual acuity (VA) measurements depend on subjective responses, which can be unreliable, especially with uncooperative participants. Objective measurements with visual evoked potentials (VEP) address this issue but can overestimate VA in amblyopia. This study aims to establish the P300 component of the event-related potential as an objective VA test for amblyopia and compare its performance to subjective (psychophysical) and VEP-based VA estimates.

Methods: Psychophysical, VEP-based, and P300-based VA estimates were obtained for amblyopic and fellow eyes of 18 participants (aged 19–65) in a bicentric study. VEP-based VA was determined from the spatial frequency threshold derived from occipital cortex pattern-pulse responses to check-sizes ranging from 0.048° to 8.95°. P300 responses were collected using visual oddball sequences with circular optotypes. The threshold was estimated from the sigmoid function of parietal P300 amplitude versus optotype gap size. Mean VA values for amblyopic eyes were compared across methods.

Results: VEP-based VA of the amblyopic eyes overestimated psychophysical VA by 0.18 ± 0.06 logMAR (P = 0.0016). In contrast, P300-based VA showed no significant difference from psychophysical VA (0.00 ± 0.04 logMAR, P > 0.05).

Conclusions: In amblyopia, P300-based optotype VA aligns more closely with psychophysical VA than VEP-based VA, suggesting that P300-based VA is a valid objective alternative for estimating VA in amblyopic eyes.

Translational Relevance: This study highlights the potential of P300-based VA testing as a reliable and objective method for assessing VA in amblyopic eyes, offering a promising tool for clinical and research applications where traditional methods fall short.

Introduction

Visual acuity (VA) testing is the most frequently performed ophthalmological examination. Its results are often of eminent importance for diagnostic and medicolegal decisions and for clinical and basic research progress. Importantly, standard psychophysical tests depend on the patient's subjective responses. Reliable cooperation is thus essential for a valid test result. However, some patients may be unable or unwilling to cooperate. Alternative approaches for acuity testing which require less cooperation are critical in such cases. Most typically, such objective acuity measures are estimated by recording visual evoked potentials (VEPs) in response to patterns, such as checkerboards or gratings, of varying coarseness.¹ If the visual system can resolve the pattern, a response is evoked—otherwise, there is no response. The pattern's coarseness at the transition from resolving the pattern to not resolving it provides an indication of the patient's acuity. Different variants of this general approach are in use and generally yield comparable results as reviewed by Hamilton et al.² 

VEP-based objective measurement of VA has proven valuable in clinical practice, covers a large acuity range,³ and is even robust against confounders such as nystagmus.⁴ However, it is critically limited by the misestimation of acuity in certain patient groups. Specifically, acuity overestimation has been reported for amblyopia.⁵ There are probably two reasons for this. First, as amblyopia is primarily a problem of cortical origin, the incoming visual information in the primary visual cortex may be largely preserved and likely able of giving rise to some VEP response irrespective of further cortical processing. Second, perception in amblyopia is distorted and fragmented,^6–8 which affects optotype recognition whereas distorted and fragmented patterns can still evoke a VEP response. Acuity overestimation with VEPs is likely a common problem for all visual impairments where vision is distorted and fragmented, as implied by experiments with artificially degraded vision.⁹ A solution to both problems is provided by the application of cognitive event-related potentials, such as the P300 component.^10,11 To this effect, preliminary evidence comes from P300 recordings to artificially distorted stimuli.¹² In contrast to VEP responses, which can only be evoked with a sufficient signal-to-noise ratio by rather massive stimuli such as checkerboards or gratings, quite subtle stimulus differences are sufficient to elicit a P300. This allows for the use of single optotypes as stimuli,¹³ which makes objective testing more akin to subjective testing. 

Taken together, although the VEP tends to overestimate acuity in amblyopia, previous developments to utilize the P300 component for objective VA estimation promise to yield more accurate measurements in amblyopia. Therefore the purpose of this study was to compare psychophysical acuity (VA_psych), VEP acuity (VA_VEP), and P300 acuity (VA_P300) from amblyopic eyes. 

Methods

Participants

Eighteen participants (10 male; 8 female; age range 19 to 65 years) with amblyopia were recruited from two centers—Department of Ophthalmology, Otto-von-Guericke-University, Magdeburg (OVGU) and University of Freiburg Eye Center (UFR). The participants were recruited through the existing hospital database and public advertisement. The screening included best corrected VA and anterior and posterior segment examination. Inclusion required a minimum interocular difference of two lines from the 4 m ETDRS chart or 0.2 logMAR determined using the Freiburg Visual Acuity and Contrast Test (FrACT).¹⁴ The fellow eye had normal or corrected-to-normal FrACT VA ranging from −0.24 to 0.16 logMAR (Table S1, column “VA_psych”). The etiologies of amblyopia in the participants were anisometropic (n = 5), deprivational (n = 4), strabismic (n = 5), mixed (n = 3), and unknown (n = 1). The three VA types were measured in two different sessions (≤30 days apart), session one comprising VA_psych (obtained with FrACT) and VA_VEP, session two VA_P300. Supplementary Table S1 in the supplementary materials shows the relevant data for each participant. Best refractive correction was determined during the first session and used for all measurements. The study followed the tenets of the Declaration of Helsinki and was part of a series of investigations with ethical approval by the institutional review boards. All participants provided written informed consent. 

Stimulation and Procedure

Subjective Visual Acuity Assessment

The VA_psych was measured monocularly using FrACT¹⁴ running on an Apple Mac Mini and displayed on a 55″ LG OLED screen (Model OLED55C17 [LG Electronics, Seoul, South Korea] at OVGU and OLED55B7D [LG Electronics] at UFR) with 60 Hz refresh rate at 3 m. The test encompassed 24 trials in total with single Landolt-C-based optotypes at eight different orientations. The measurement was done twice for each eye using an “ABBA” scheme with the right eye always first. 

Visual Evoked Potential-Based Acuity Assessment

Stimuli and Task

The participants were presented with 14 different checker sizes that were exponentially equidistant from 0.05° to 8.95° in a brief-onset mode (40 ms on, 93 ms off) at 7.5 Hz (steady-state VEP) on a CRT monitor (FIMI-Philips, Saronno, Italy, at UFR; MDG403 [Philips, Amsterdam, the Netherlands] at OVGU). The room was dimly illuminated and the observation distance was 114 cm for six smaller checker sizes (0.05°–0.38°) and 57 cm for eight larger checker sizes (0.52°–8.95°).¹⁵ Based on the observation distance, the recording was split into two different measurement blocks. Space-averaged mean luminance was 45 cd/m² with 40% contrast, in agreement with the respective ISCEV Extended Protocol.¹ A total of 10 sweeps of each checker size was presented for 1067 ms in a repetitive sequence. For each checker size, over six cycles, 60 sweeps or 480 onsets were performed. The participants were instructed to report random numerals that appeared in the center of the screen at random intervals. This task ensured the participants’ fixation, attention and accommodation. VEPs were recorded and analyzed following the methodology established by Bach et al.¹⁵, as detailed in the supplement. 

P300-Based Visual Acuity Assessment

Stimulation and data analysis for P300-based acuity estimation was distinctly different from that for VEP-based estimation. 

Stimuli and Task

Visual stimuli were generated using PsychoPy 2¹⁶ on an Apple Mac Mini and displayed on the same screen as used for psychophysical testing with 60 Hz refresh rate at a distance of 3 m. The screen was the only source of illumination. To elicit P300, we employed an oddball paradigm, using the “FreiBurger” optotype, which is a filled black circle with the critical detail of a gap across the full diameter, oriented at ±45° relative to the vertical axis, extending through the center as rare stimuli (Fig. 1B, bottom graph). It was interleaved by the frequent filled black circles without a gap.¹⁷ The participants indicated the orientation of the gap via a button box. On average, every seventh stimulus was a rare one (i.e., a FreiBurger optotype), yielding a sufficient signal-to-noise ratio.¹⁸ The oddball sequence was fast, with an interstimulus-onset interval of 216 ms, coupled with a jitter of 250 ms. This timing, including a stimulus presence of 100 ms and an inter-stimulus interval of 116 ms, closely resembled the parameters employed by Mell et al.¹⁹ and allowed for the extraction of transient event-related potentials despite the rapid stimulation. The stimuli, with a Weber contrast of 99%, were presented on a white background with a luminance of 130 cd/m², following established standards for optotype presentation.²⁰ Interstimulus interval was a blank screen with the same luminance. The stimuli were presented in six different sizes, distributed around the monocular VA_psych, ranging from a gap size of −0.15 log arcmin below the participant's per-eye thresholds to 0.60 log arcmin in steps of 0.15 log arcmin. These relative gap sizes are denoted as “ΔGap size” in Figure 1B (identical for the amblyopic and the fellow eye) and converted to absolute gap sizes in the bottom graph of Figure 1B (resulting in different values for both eyes). 

Figure 1.

 

Data and analysis of a representative participant for VEP-based (A) and P300-based VA estimation (B). Psychophysical acuity (VA_psych) for the amblyopic and the fellow eye was 0.53 and −0.18 logMAR respectively. (A) VEP-based acuity estimates (VA_VEP) from 14 different check sizes. Original traces are provided in Supplementary Figure S1. The tuning curve demonstrates VA_VEP to overestimate VA_psych for the amblyopic eye (0.08 vs. 0.53) and to be in closer agreement for the fellow eye (0.04 vs. −0.18). (B) Event-related potential traces from the odd-ball paradigm for P300 recording for 6 different relative gap sizes (ΔGap size). P300 tuning curves (bottom) show the sigmoid dependence of amplitudes on absolute gap size for both amblyopic and fellow eye. Raw P300-based acuity estimates (VA_{P300_raw}) are obtained from the inflection points of the curves, similar to psychophysical thresholds.

Event-Related Potential Recording and Analysis

Event related potentials (ERPs) were recorded using a 32-channel EEG system (Brainamp) with active electrodes (Acticap; Brain Products, Gilching, Germany) and software (Brain Vision Recorder 2.0; Brain Products). The electrodes were mounted according to a subset of the international 10–10 system,²¹ initially referenced to the FCz. FPz served as the ground electrode. A vertical EOG was recorded from surface electrodes (Ternimed; Wolfram drool GmbH, Mainz, Germany) placed 1.5 cm above and below the eye. Overall, the impedance was kept below 5 kΩ at each electrode. The signals were bandpassed at 0.1–70 Hz, digitized at a sampling rate of 500 Hz, and subsequently saved to storage. 

Data analysis was performed in IGOR Pro 8 (Wavemetrics Inc., Lake Oswego, USA). The recordings were re-referenced to linked mastoids (TP9/TP10)¹³ and trials were pooled based on stimulus size irrespective of orientation, isolated for each rare and frequent stimulus. Before averaging, the signals were low-passed at 5 Hz (acausal zero-phase-shift FFT-based filter). Artifact rejection was based on a threshold criterion of ±200 µV applied to the interval of −200 ms to 900 ms around stimulus onset. Data segments were cut from −100 ms to 800 ms around stimulus onset and a trial-by-trail baseline correction was applied based on the interval ranging from −100 ms to 50 ms around stimulus onset. That way around 120 rare and 720 frequent artifact-free trials were averaged to yield ERP curves. To isolate responses related specifically to the rare stimuli, we computed difference traces by subtracting the ERPs to frequent stimuli from the ERPs to rare stimuli. The P300 was isolated at Pz, where it typically reaches its maximum.²² To obtain the amplitude, a time window ranging from 250 ms to 700 ms after stimulus onsets was defined, in which the P300 is expected to peak.^22,23 This concept was successfully applied in previous studies.^19,24 

Being based on cognitive responses, P300-based acuity estimation follows similar basic principles as psychophysical testing. Based on the amplitudes of the responses to the six presented stimulus sizes a sigmoid function, resembling a psychometric function, was fitted and VA_{P300_raw} was estimated from the inflection point. However, despite their conceptual similarity, VA_{P300_raw} and the psychophysical acuity value cannot be expected to be numerically equal, given, for example, inherent procedural differences including different temporal characteristics. To convert VA_{P300_raw} to an estimate of psychophysical acuity for the amblyopic eye, a conversion parameter was required. Two variants were established, an individualized and a mean-based conversion parameter yielding VA_{P300_iConv} and VA_{P300_mConv}, respectively (Fig. 2): 

Individualized conversion: 

• Step 1: Δ_iConv = VA_{P300_raw_fellow eye} − VA_{psych_fellow eye}
• Step 2: VA_{P300_iConv} = VA_{P300_raw_amblyopic eye} − Δ_iConv

Mean based conversion: 

• Step 1: Δ_mConv = mean(Δ_iConv)
• Step 2: VA_{P300_mConv} = VA_{P300_raw_amblyopic eye} − Δ_mConv

Figure 2.

 

Relation of VA_P300 and VA_psych for all participants (n = 18) and effect of conversion parameters. (A) VA_{P300_raw} vs VA_psych for fellow and amblyopic eyes. For the determination of VA_{P300_raw} from the tuning curve see Figure 1B and text. (B) VA_{P300_iConv} vs VA_psych for the amblyopic eyes. VA_{P300_iConv} is based on each individual's fellow eye difference between VA_{P300_raw} and VA_psych (i.e., conversion parameter, as detailed in text). (C) VA_{P300_mConv} versus VA_psych for the amblyopic eyes. VA_{P300_mConv} is based on the group's mean conversion parameter across all individuals’ fellow eyes, as detailed in text. (D) Comparison of two conversion methods (individual and mean-based) from B and C. Difference of VA_{P300_iConv} and VA_psych versus VA_{P300_mConv} and VA_psych. The orange circle with black filling represents the mean and 95% confidence interval. As no systematic shift from the identity line is evident, both methods are equally effective for VA_P300 conversion. Note the different scales used for D.

Statistics

To determine 95% confidence intervals of the mean VAs where applicable, a non-parametric bootstrap with 10,000 repetitions was performed in IGOR Pro 8 (Wavemetrics Inc., Lake Oswego, OR, USA). To assess the statistical significance of the comparisons between VA measures, paired t tests were performed with R (R: Foundation for Statistical Computing, Vienna, Austria). Bland-Altman plots were generated and limits of agreement were determined with R. 

Results

The data from a representative participant are depicted in Figure 1. Here the VA estimation from VEPs (i.e., VA_VEP) and P300-ERPs (i.e., VA_{P300_raw}) are juxtaposed for the fellow and amblyopic eye (for raw VEP traces see Supplementary Fig. S1). The VA_VEP estimates (Fig. 1A) were derived according to Bach et al.¹⁵ (i.e., with a conversion parameter that determines the VA_VEP from the spatial frequency limit of the VEP responses). It demonstrates the previously reported overestimation of VA_VEP⁵ for the amblyopic eye (0.08 vs. 0.53 logMAR). In contrast, the match is much closer for the fellow eye (i.e., 0.22 logMAR [0.04 vs. −0.18 logMAR]). Taken together, although VA_psych differs considerably between fellow and amblyopic eye, VA_VEP are similar between both eyes. The VA_{P300_raw} estimates were derived from the gap-size tuning of the P300 amplitude as detailed in Methods. VA_{P300_raw} is clearly worse for the amblyopic than for the fellow eye (0.74 ± 0.03 [CI] vs. 0.00 ± 0.05 log MAR), which is in general agreement with the intraocular VA_psych difference (0.53 vs. −0.18 logMAR). It is also evident that, for both eyes, VA_{P300_raw} and VA_psych differ by at least 0.2 logMAR. This is expected as, in contrast to the above VA_VEP, an appropriate conversion parameter has not been applied to convert VA_{P300_raw} to VA_{P300_conv}. As described below for the group analysis and in Methods, we determined the conversion parameter from the fellow eyes to obtain VA_{P300_mConv} for the amblyopic eye. 

Group Results for VA_psych and VA_P300

The trend observed for the above individuals also holds for the entire group as reflected by the group results depicted in Figure 2. As Figure 2A shows, VA_{P300_raw} is consistently worse (higher logMAR) than VA_psych. As indicated above, to test the actual correspondence of VA_{P300_raw} and VA_psych, a conversion parameter must be applied to yield VA_{P300_Conv}. This conversion parameter was obtained from the fellow eye's data and applied to the amblyopic eye's data to yield VA_{P300_Conv}. This can be done either intraindividually, using each individual's own fellow eye, to yield VA_{P300_iConv} (“individual conversion”, see Fig. 2B), or based on the mean data across all individuals' fellow eyes to yield VA_{P300_mConv} (“mean-based conversion”, see Fig. 2C). Both methods result in a similarly close match of VA_{P300_Conv} and VA_psych. This is directly assessed in Figure 2D, where the difference between VA_{P300_Conv} and VA_psych is compared for the individual and mean-based conversion. Calculation of the mean revealed similar results for both conversion parameters (difference for VA_{P300_iConv}: Mean = 0.01, CI = −0.044 to 0.064; difference for VA_{P300_mConv}: Mean = 0.01, CI = −0.042 to 0.067). VA_{P300_mConv} was used for the subsequent analyses, as it is a more general approach also applicable for use in other cohorts. The conversion parameter to obtain VA_{P300_mConv} was 0.23 logMAR. 

Group Results for VA_{P300_mConv} and VA_VEP

The depiction of VA_{P300_mConv} and VA_VEP vs VA_psych (Fig. 3) indicates a closer match with VA_psych for VA_{P300_mConv} than for VA_VEP. This is more formally analyzed in the Table. Importantly, VA_{P300_mConv} does not differ significantly from VA_psych, whereas VA_VEP does strongly differ for the amblyopic eye (P = 0.0016). There is also a weakly significant effect for the fellow eye (P < 0.03). Finally, we assessed the limits of agreement (LOA) between the electrophysiological measures and the VA_psych (Supplementary Fig. S2) with focus on the LOA for VA_{P300_mConv} for the amblyopic eye, where we obtained LOA of ±0.24. As reference, the LOA for VA_VEP (for the valid estimates of the non-amblyopic fellow eye) is ±0.26. This indicates similar confidence intervals for both electrophysiological approaches. 

Figure 3.

 

VA_VEP and VA_{P300_mConv} vs VA_psych in amblyopia. Individual data points (n=18) for VA_{P300_mConv} more closely match the identity line than those for VA_VEP. Mean VA_{P300_mConv} largely coincides with the identity line, indicating a close agreement with VA_psych, while the mean of VA_VEP deviates towards the ordinate, indicating an overestimation, by on an average exceeding one line of visual acuity (0.18 logMAR) using VEP. Some participants show a considerably larger deviation. Unfilled symbols represent the respective means with 95% confidence interval.

Table.

 

Comparison of VA Estimates at the Group Level (n = 18)

 

Discussion

We report that objective VA estimates in amblyopia are more accurate with P300 than with VEP measurements. Thus P300-based acuity estimation with optotypes may be of promise to fill a gap for objective VA assessments, especially in specific conditions of VA_VEP misestimation. 

Comparison With Previous Findings

In accordance with previous reports,^2,5,25 we observed an overestimation of VA with VEPs (i.e., VA_VEP). The match of VA_P300 with VA_psych in amblyopia has not been investigated until now. Our observation of a closer match with VA_psych for VA_P300 than for VA_VEP in amblyopia, however, is expected from previous P300 research with imitated amblyopia.¹² In fact, this can be predicted from psychophysical studies which demonstrated a discrepancy between grating acuity and optotype acuity in amblyopia, that resembles the overestimation of acuity in VEP-based estimates.^26–29 To relate objective VA measures to VA_psych, a conversion parameter is recommended.² For VA_P300 we determined a conversion parameter of 0.23 logMAR, which slightly differs from the parameter determined in a pilot study¹³ (i.e., 0.36 logMAR) where dioptric blur was used to degrade vision. 

Mechanisms and Value in Basic Research

Using the cognitive P300 component with optotypes for estimating VA is likely much akin to psychophysical acuity testing, as both involve cognitive processing. Deriving the P300-based threshold value from a sigmoid curve¹³ accounts for the P300 amplitude value not reaching zero even below threshold, unlike VEP where an extrapolation approach is used. The conceptual similarity between the P300-based approach and psychophysics raises the question why a conversion parameter is required at all. At least part of the answer is likely to lie in temporal aspects of optotype recognition. For a distinct P300 to be elicited, reliable stimulus recognition based on the initial, say, 100–200 ms of optotype presentation is required. This is different from standard psychophysical acuity procedures where the duration of optotype inspection is not limited. It is known from a number of psychophysical studies that the outcome of an acuity test is adversely affected by shortening the duration of optotype presentation. For instance, Heinrich et al.¹⁰ found a difference of 0.29 logMAR between 1.0 s and 0.1 s presentation times. Because P300 generation does not depend on a specific type of visual stimulus, this method can potentially be extended to evaluate other aspects of visual performance including crowding. On the other hand, the same underlying mechanisms that cause VA_VEP to overestimate VA_psych in amblyopia might prove useful to predict post-therapy acuity as reported by Ridder and Rouse.³⁰ 

Transfer to Patient Applications

To our knowledge, the present study is the first to report P300-based VA estimates in a clinical population. The performance of P300-based acuity estimates in amblyopia thus serves as proof-of-principle for this approach in conditions where VA_VEP is systematically misestimated. Beyond amblyopia, this might include other conditions where grating acuity and optotype acuity diverge, as is the case with optical aberrations,³¹ eccentric viewing,³² cerebral visual impairments,³³ in preschool children,³⁴ and likely in cases of metamorphopsia. 

The potential of P300-based acuity estimation for clinical investigation motivates further developments to transform the current paradigm into a clinically useful test. This should specifically address the time needed for an acuity estimation. This issue could potentially be addressed in multiple ways. On the one hand, one could reduce the number of optotype sizes to which the P300 is recorded. Depending on the specific question, a single optotype size might be sufficient to assess whether a patient's acuity is above or below a certain limit. On the other hand, it might be possible to use stimulation sequences to potentially make use of frequency tagging, where responses can be quantified efficiently using frequency-space analysis.³⁵ 

Limitations and Outlook

Besides the practical issues discussed above, future studies should address potential limitations of our present study. These include the conversion of the P300 threshold value to the scale of psychophysical VA measurements. Here we used the fellow-eyes of the amblyopic study participants to determine the conversion parameter. Although it is reassuring that the outcome is nearly identical when individual values are used as conversion parameters instead of the group-based parameter, the general applicability of this value could be further validated in a sample of control participants. The results of this study might have been even clearer if we had looked at a single sub-entity of amblyopia (strabismic, deprivational, or anisometropic) and separately assessed different levels of acuity. An interesting aspect would be the effect of amblyopia treatment on the discrepancy between VEP-based and psychophysical acuity. In conclusion, we see our findings as a promising result, suggesting a strong role for P300-based optotype VA estimates as a complement in the arsenal of objective VA measures for both clinical applications and basic research. 

Acknowledgments

The authors gratefully acknowledge the support by the study participants and Enyam Komla Amewuho Morny and Ms. Laura Dietrich for their contributions during the initial part of the study. 

Supported by the German Research Foundation to MBH and SPH (DFG; 435838478, HO 2002/22–1 & HE 3504/9–1) and by the European Union's Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No 955590 (OptiVisT) to MBH. 

Disclosure: A.V. Gopiswaminathan, None; J. Haldina, None; K.O. Al-Nosairy, None; C.Z. Duval, None; F.H. Stolle, None; M.B. Hoffmann, None; S.P. Heinrich, None