Pattern-reversal visual evoked potentials in prosthetic vision and simulated visual reduction

Participants

The Argus II group consisted of four male participants who were blind due to end-stage retinitis pigmentosa and were treated with the Argus II retinal implant (mean age=66.8 years, range=60–76 years; table 1). Additionally, a convenience sample of seven sighted control participants was tested (mean age=41.4 years, range=26–62 years). Participants provided written consent prior to study participation. Argus II patients additionally signed a media consent form, allowing the release of their photo and video recordings for research and educational purposes.

Stimuli

We used the JÖRVEC Visual Pattern Stimulator (JÖRVEC, Miami, Florida, USA) to present visual patterns (figure 1A). The display’s on-demand refresh scheme ensures a high pattern update speed of less than 100 µs, preventing aliasing issues that could arise from the mismatch between the frame rate of the implant camera (30 Hz) and the screen refresh rate of conventional displays (typically ~60–120 Hz). However, due to the inability to directly access the camera’s video synchronisation signal, there was an uncertainty in the onset of electrically elicited VEP (eVEP) responses for the Argus II participants, including a 0–33.3 ms variable delay due to camera frame update and the processing time to convert video signals to electrical signals. The impact will be discussed later.

Request permission

The stimuli consisted of four interleaved black and white horizontal stripes (Michelson contrast=98.8%). Stimuli luminance was measured using a photometer (CS-100; Minolta, Tokyo, Japan). Argus II users viewed the pattern using their prostheses while the experimenters monitored the camera input via the Argus II Clinician Programming System laptop (figure 1A). The camera’s field of view was an oblique rectangle of 11 by 18 degrees, and the viewing distance was adjusted to fully enclose the top and the bottom edges of the pattern in the camera’s view (figure 1A; online supplemental video). The resulting pattern size was approximately 20 degrees or 0.1 cycle/degree (cpd) vertically. Sighted controls viewed the pattern from 40 cm away to match this spatial frequency, with a small red fixation dot affixed to the centre of the screen (figure 1). Since the stripes are slanted relative to the implant, some electrodes near the edges of the bars might be rendered as grey pixels rather than black or white, reducing the sharpness of the edges (see online supplemental video).

For Argus II users, the dark and light stripes reversed at 1.01 reversals/second (rps) to allow sufficient time for the VEP to return to baseline. Controls experienced a higher reversal rate of 3.37 rps to maximise trial and participant testing within a short time frame due to the limited availability of participants, experimenters and equipment. While a higher reversal rate may alter VEP shape and reduce peak amplitude, at 3 rps, the response remains transient rather than a steady-state response.10 Additionally, our inferential statistical analyses focused solely on within-subject comparisons in the control group, minimising the influence of this difference in reversal rate.

Viewing conditions

The Argus II users viewed the pattern through the camera with both eyes patched to eliminate residual light response. Controls viewed the pattern using their best-corrected vision in the right eye only, as the implant is placed monocularly. An exception is Control 02 in the Normal Vision condition, which was tested binocularly by error. This binocular data set is reported in graphs and tables for completeness but is excluded from statistical analyses.

The Argus II users were only tested under the Prosthetic Vision condition, viewing the full-field, full-contrast pattern using their implant without any degradation. The room was well-lit to minimise the influence of the automatic gain control of the device. The control group viewed the pattern under five conditions (figure 1B): (1) Normal Vision: viewing the full-field, full-contrast pattern without alterations. (2) Restricted Field: viewing the same pattern with simulated field restriction. (3) Blurry Vision: viewing the pattern with both field restriction and through a blurry filter. (4/5) Reduced Contrast: viewing the clear pattern at 32% and 6% of the full display contrast, with field restriction.

To simulate field restriction, a modified pair of welding goggles was used to occlude both eyes, leaving only a small oblique rectangular opening in the right eye, matching the typical video’s field of view of 11×18° in Argus II. Blurry vision was simulated by attaching Bangerter Occlusion Foils11 to a pair of flip-up clear acrylic lenses mounted on the goggles. This reduces visual acuity to approximately 20/90012 and Michelson contrast to 35.2% as measured by the photometer. Since the fixation dot on the screen was not visible once blurred, participants first adjusted the field of view without the blur to centre the pattern and then put down the blurring lens for the recording. In addition, a fixation dot was drawn on the blurring lens in the centre of the rectangular viewing window to help with maintaining fixation. Display contrast reduction was achieved by adjusting grey levels of the stripes while maintaining the overall luminance level. A contrast reduction of 32% was selected to approximate the Michelson contrast in the Blurry Vision condition (35.2%), while 6% was selected to evaluate the responses when the stimuli were barely visible. These conditions were designed to simulate the effect of degraded visual input rather than replicate the perception of Argus II users.

VEP recording

VEP responses were recorded using the JÖRVEC Visual Pattern Stimulator (JÖRVEC) and the Universal Smart Box acquisition system (Intelligent Hearing Systems, Miami, Florida, USA). The electrode montage followed the International Society for Clinical Electrophysiology of Vision (ISCEV) standard,13 with the active electrode at Oz, the reference at Fz and the ground over the forehead. A typical pattern-reversal VEP has an initial negative peak at approximately 75 ms (N75), followed by a positive peak at around 100 ms (P100) and another negative peak at around 135 ms (N135).13 P100 is proposed by many researchers to originate from V1/V2, with the relative contribution of the two areas depending on stimulus size.14 15 Therefore, our study uses the peak amplitude and latency of P100 as our primary outcome measure to assess the preserved function of early visual areas in AVTS users.

Responses were collected at 1.01 rps for Argus II subjects and 3.37 rps for controls (see ‘Stimuli’ section above), with 512 samples being captured between each reversal. Recordings automatically stopped when the predetermined number of reversals, typically 256 or 200, was attained. The average recording time for a 200-reversal session was 3.5 min for the Argus II subjects (or 6 min including setup) and 1 min for controls (or 2 min including setup). For Argus II participants, a minimum of three recordings were collected, and additional recordings were conducted until the quality of VEP responses could not be further improved. For controls, three recordings (600 reversals) were collected per condition, except for the first participant where six recordings were collected to verify the reproducibility of the recordings. The total number of reversals collected for each participant in each condition, as well as P100 latency and amplitude, are summarised in table 2.

VEP data processing

During acquisition, data were filtered between 1 Hz and 300 Hz, and trials with eye movement or blink artefacts were automatically discarded (rejection rate=20.6% for Argus II participants and 12.4% for controls). While nystagmus has been noted in Argus II patients (eg,16), it has minimal influence on our results because the video input is based on the position of the camera rather than the eye, and the trials with oculomotor artefacts have been removed. A fourth-order Butterworth frequency filter between 0.3 Hz and 70 Hz was applied prior to analysis. P100 latency was defined as the time between each reversal onset and the peak of P100. P100 amplitude was the difference in microvolts (μV) between the preceding negative peak, N75, and the peak of P100. Reversal onset was set as time zero, and the peak of N75 was set to 0 μV as the baseline. All reversals for each participant, under each condition, were epoched between 0 and 250 ms, aligned and averaged to generate individual plots.

Statistical analyses

We used R (V.4.2.0) and RStudio (V.2022.12.0+353) for statistical analyses and data plot generation. Given the limited sample size and significant inter-subject variability in prosthetic VEPs, only descriptive statistics for the Argus II group are reported. For the control group, we employed linear mixed-effect (LME) models to evaluate the impact of degraded vision on VEP responses using the ‘nlme’ package.17 Fixed effects included field (normal/restricted), vision (clear/blurry) and contrast (full/reduced 32%/reduced 6%). Variations across participants were modelled as random effects. We fitted separate models for P100 latency and log-transformed P100 amplitude. Type-II analysis of variance tests were conducted on these models to summarise the fixed effects, with reported χ², p values and 95% CIs of group mean or difference values estimated by the LME models. Differences in log-transformed amplitude were expressed as percent changes for clarity.

Prosthetic vision versus normal vision

Figure 2 compares the VEP responses of Argus II patients (black) and control subjects under the Normal Vision condition (orange) when they both viewed the full-contrast stimuli without additional degradation. The prosthetic VEPs in Argus II participants exhibited a major positive peak, preceded and followed by two negative peaks, resembling the typical VEP pattern observed in control subjects. This demonstrates that the visual cortex of Argus II patients remains responsive to artificial retinal input and exhibits a comparable response profile to controls. However, there was greater variability in P100 latency and amplitude (table 2). A further examination of the odd and even sweeps indicates that the waveforms in Argus 02–04 are reliable and likely reflect their true brain activity, whereas the waveform in Argus 01 may be largely driven by noise (online supplemental figure 1).

Request permission

Artificial vision reduction

The input of the Argus II retinal prosthesis differs from typical visual input in several key aspects. To examine the impact of degraded visual input on VEPs, we fitted LME models where Field (full/restricted), Vision (clear/blurry) and Contrast (full/32%/6%) were included as fixed effects and individual differences were modelled as random effects. Figure 3 summarises these comparisons. For clarity, each panel only included participants who were tested under all the conditions for that comparison. However, all available data were included in the LME models, and the three factors were evaluated in the same model rather than being tested individually.

Request permission

The LME models reveal that blur significantly delayed P100 peak time by 8.7 ms (95% CI (0.9, 16.6), χ² (1, N=30) = 5.38, p=0.02), while reducing contrast significantly decreased P100 amplitude (χ² (2, N=30) = 57.36, p<0.001; at 32% and 6% contrast, P100 amplitude decreased to 55% (37%, 82%) and 20% (13%, 31%) compared with full-contrast, respectively). No other factors had statistically significant effects.

Measuring VEP responses in AVTS users

To our knowledge, the current study is the first to report VEP responses in human retinal prostheses users actively using their device to view a pattern reversal stimulus. Previous eVEP recordings in humans were obtained by directly delivering brief pulses to the microelectrodes using a computer programme.6 7 18 This is analogous to flash VEPs, which are known to be more variable in morphology and timing compared with pattern-reversal VEPs.13 Compared with previous methods, our approach provides a holistic evaluation of the actual usage of the device and enables direct comparison between prostheses users and controls using the same visual stimuli. We found that the pattern-reversal VEPs using prosthetic vision and normal vision exhibit similar morphology. Consistent with previous studies, reproducible eVEPs in prosthesis users could be recorded, although the waveforms exhibit high variability across participants.6 7 18 Together, these results highlight the promise of using visual prostheses as a treatment for blindness.

Brain plasticity after losing and regaining sight

The similarity in morphology between prosthetic and typical VEPs suggests that the visual cortical function of AVTS users is, to some extent, preserved. Consistently, brain imaging studies have provided evidence of retained retinotopic mapping in participants with adult-onset blindness following sight recovery surgeries.19 20 Furthermore, increased use of AVTS after implant surgery can enhance vision-induced activity in the visual cortex21 and reduce cross-modal activation.22 Nevertheless, it is crucial to recognise that prior experience with form vision may be essential for the recovery of complex visual functions.23–26 All our participants had light perception before implantation, had been using their device for 0.5–2 years prior to the experiment and had formed vision prior to receiving Argus II implantation, likely contributing to the preservation and recovery of their visual cortical functions.

Factors that may affect prosthetic VEPs

Because prosthetic vision differs from natural vision in both its input and the perceptual experience, it is important to consider the factors that can influence the waveform of prosthetic VEPs. The findings from the current study, as well as previous research, have consistently shown that blur increases P100 latency27 28 and reduced contrast or contrast sensitivity decreases P100 amplitude.27 29 30 For Argus II users, their perceived clarity and contrast are influenced by both the camera’s video input and the neuronal responses to stimulation. Existing AVTSs may have already implemented contrast enhancement and automatic gain control algorithms, but patients with severe RP may experience transsynaptic degeneration of retinal cells,31 limiting the electrically elicited neural responses. Moreover, for an epiretinal implant like the Argus II, activating one electrode could stimulate a bundle of ganglion cell axons from both local and distant neurons, resulting in a blurred perception of a line,32 33 reducing the perceived clarity and contrast of prosthetic vision.

For AVTS users, their P100 latency is influenced by additional factors. In the case of Argus II, due to the slow refresh rate of its camera (30 Hz), there existed a variable delay (0–33.3 ms) between the image onset on the pattern stimulator screen and the electrode stimulation on the retina, delaying and reducing the recorded P100 peak (see online supplemental figure 2 for simulated effects). On the contrary, since the electrodes bypass a portion of retinal processing and directly stimulate ganglion cell axons, eVEP could have a shorter latency compared with light-elicited VEPs. This phenomenon has been reported in animal models with subretinal implants34 and analogously in cochlear implant users.35 Therefore, it is crucial to interpret the latency of prosthetic VEPs within the context of these factors.

Lastly, the neural thresholds associated with each electrode could vary greatly. In our sample, the thresholds typically vary by more than fourfold within a participant. The responses may be mostly driven by the electrodes with lower thresholds, similar to the observation of Stronks et al.7

Clinical applications

Measuring VEPs offers an objective means to assess AVTSs, including but not limited to the Argus II. Similar practices have been adopted for cochlear implants both preoperatively to aid in assessing patient eligibility36 and postoperatively to objectively evaluate surgical outcomes, optimise implant fitting and improve speech outcomes.37 Our approach requires solely a pattern-reversal VEP device with a sufficiently high pattern update speed, which many clinics may already possess, and can be applied universally to patients receiving any type of visual prostheses. Recommended practices for measuring prosthetic VEPs include (1) patching both eyes of the patient to eliminate residual vision, (2) using luminance-balanced patterns in well-lit environments to minimise the impact of the device’s automatic gain control, (3) selecting a low reversal rate to ensure return-to-baseline time and avoid aliasing and (4) comparing the odd and even sweeps to ensure data quality. Limitations include the uncertainty in the exact location and timing of the electrode stimulation on the retina and the coarseness of the pattern compared with the clinical evaluation of normal vision. Future studies could explore ways to individually control the electrodes with precise timing using computer programmes and deliver customised patterns based on the placement of the implant. This method could be used to collectively establish a database with diverse samples and longitudinal data, enabling more informative prognoses.