Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2014 Nov 4:8:816.
doi: 10.3389/fnhum.2014.00816. eCollection 2014.

Simulation of thalamic prosthetic vision: reading accuracy, speed, and acuity in sighted humans

Milena Vurro  1 , Anne Marie Crowell  1 , John S Pezaris  1
Affiliations

Simulation of thalamic prosthetic vision: reading accuracy, speed, and acuity in sighted humans

Milena Vurro et al. Front Hum Neurosci. .

Abstract

The psychophysics of reading with artificial sight has received increasing attention as visual prostheses are becoming a real possibility to restore useful function to the blind through the coarse, pseudo-pixelized vision they generate. Studies to date have focused on simulating retinal and cortical prostheses; here we extend that work to report on thalamic designs. This study examined the reading performance of normally sighted human subjects using a simulation of three thalamic visual prostheses that varied in phosphene count, to help understand the level of functional ability afforded by thalamic designs in a task of daily living. Reading accuracy, reading speed, and reading acuity of 20 subjects were measured as a function of letter size, using a task based on the MNREAD chart. Results showed that fluid reading was feasible with appropriate combinations of letter size and phosphene count, and performance degraded smoothly as font size was decreased, with an approximate doubling of phosphene count resulting in an increase of 0.2 logMAR in acuity. Results here were consistent with previous results from our laboratory. Results were also consistent with those from the literature, despite using naive subjects who were not trained on the simulator, in contrast to other reports.

Keywords: artificial sight; neuroprosthesis; neurotechnology; visual prosthesis.

PubMed Disclaimer

Figures

Figure 1
Figure 1
System architecture. The system consists of the Tobii TX300 gaze tracker operating in normal, streaming mode. Gaze position information relative to the TX300's built-in display screen is streamed over a low-latency dedicated connection to an intermediary interface computer that runs a small gaze-position server program. The server code accepts streaming data and, upon request from the behavioral control system, computes an instantaneous gaze position value (with non-linear noise reduction), that is returned over a second, low-latency dedicated connection. The behavioral computer runs the experiment and performs data logging. Whenever the behavioral task requires instantaneous gaze information (such as during the Reading Phase of the task; see main text), a request is sent to the gaze-position server that typically replies in under 2 ms. Total system latency from gaze measurement to video frame update is typically under 16 ms, with an additional 4 ms of LCD lag to an updated image on the subject retina. The intermediary gaze-position server eliminates the need for the behavioral control system to deal with streaming data, and thus simplifies the overall design, at the cost of a modest increase in total system latency.
Figure 2
Figure 2
A subject in front of the apparatus. A subject is shown seated in front of the Tobii TX300 gaze tracker, in a typical position for performing the experiment. No chin bar or other head stabilization is necessary with the TX300. The stimulus display screen in front of the subject shows a frame from the Reading Phase where a collection of phosphenes can be seen to very roughly depict three lines of large text. The bluish screen background and dark halos around the phosphenes are artifacts from the off-axis viewing angle of the camera; to the subject, the background appears uniformly black and the phosphenes as white Gaussians (see Figure 3).
Figure 3
Figure 3
A screen image taken during the reading phase of the task. Here, the simulation of thalamic artificial vision with 2000 simulated phosphenes (far fewer are engaged in this image, due to the limited extent of the display, and to the particular bright/dark pattern forming the text) is shown with the subject looking at text displayed at the largest font size. The image depicts the text “My father takes me/to school every day/in his big green car” with the point of regard alighted on the h in school. The details of image generation for phosphene vision are found in Figure 4. As the point of regard moved around the screen, the pattern of phosphenes shifted accordingly, revealing more detail wherever the subject was looking. Because of the ability of the human visual system to integrate information across eye movements, the text appeared far more legible as a whole than this static image would imply. Every subject was able to read at the depicted condition (largest text with densest phosphene pattern) with relative ease and 100% accuracy.
Figure 4
Figure 4
Real-time stimulus generation. Each frame during the Reading Phase of phosphene vision trials was generated in real time according to the flow chart shown here. The phosphene pattern selected for the trial (A), was translated in visual space so that the location (0, 0) was centered on the instantaneously read gaze position relative to the screen (B), to generate a pattern relative to the point of regard (C). The text for the given trial was then rendered as an off-screen image (D), and the translated pattern of locations overlaid on that image as a set of independent averaging filters. Each phosphene brightness was determined by taking the local average luminance of the text image weighted by a two-dimensional Gaussian filter that was sized according to the eccentricity of the phosphene (see main text). The outputs of individual filters (E) were used to set the brightness of matching-sized Gaussians at the corresponding locations. When phosphenes overlapped, they were combined additively in the final image. After all phosphenes were rendered (F), the entire frame was copied to the video card. Typical processing time for each frame was less than the refresh time of the subject monitor, so that each monitor refresh could contain a new update, and thus create a real-time simulation. Since the phosphene pattern was always translated to the point of regard, the procedure had the effect of (coarsely) stabilizing phosphene locations on the retina, matching the expected behavior of a real device (Pezaris and Reid, 2007). The frame generated here corresponds to the example shown in Figure 3.
Figure 5
Figure 5
Phases of the experimental task. One trial of the task is shown in a sequence of snapshots. The task had four distinct phases, Start, Pre-Stimulus, Reading, and End, with the bulk of the time spent in the Reading phase. Green-blue dots indicate the initial central fixation point used to engage the trial, and the subsequent dot at the center top used to advance to the next trial. Red crosses indicate the instantaneous gaze position, and red arrows, the gaze motion from one snapshot to the next (neither would appear to the subject). During the Reading phase, the subject is free to look about the screen, but gaze patterns typically followed the three lines of text with a series of fixations on each line. The (simulated) subject in this instance can be seen to read across each of the three lines, left to right and top to bottom, before looking to the trigger point in the End phase. The image in the center panel corresponds to the image in Figure 3.
Figure 6
Figure 6
Reading accuracy. Population reading accuracy vs. font size, expressed as equivalent acuity is shown. Data have been grouped by viewing condition: Clear (blue; text presented in the clear as a control condition), High density phosphene pattern (green), Medium density pattern (yellow), and Low density pattern (red). Data points are the mean, error bars are the standard deviation of population values. Smooth curves are logistic fits with assumed asymptotes of 0 and 100% for all three phosphene vision conditions.
Figure 7
Figure 7
Reading speed. Population data are shown for reading speed vs. font size (expressed as equivalent acuity), broken down by viewing condition. The upper panel shows the Clear condition where text is shown in the clear as a control (blue). The lower panel shows the results for simulated artificial vision with the three phosphene patterns, High density (green), Medium density (yellow), and Low density (red). Points are the mean, error bars are the standard deviation of population values. The two panels have different vertical scales; the control viewing condition has been separated out to display more detail in the simulation viewing conditions. Logistic fits were not performed with these data as only one condition spanned lower to upper performance plateaus as necessary for a valid fit, and we do not have direct evidence that the asymptotes are identical for all three conditions. As at least two curves would be necessary for meaningful comparisons, a fit for the High data is not shown.
Figure 8
Figure 8
Instrumentation effects. The panels in this figure depict per-subject mean values (blue unfilled circles), along with least-squares linear regression fits (red solid lines) and 95% confidence intervals (red dashed lines). (A) Reading Accuracy vs. Tracking Quality. With an R2-value of close to zero, a relationship was not found between reading accuracy αtotal (the per-subject mean over all viewing conditions and font sizes) and tracking accuracy (the percentage of valid 300 Hz samples). Removing the outlier with 76% tracking does not materially affect the result. Tracking quality can be less than 100% because either of blinks, or loss of localization of the eyes by the gaze tracker. In both cases, the gaze-position sever code provided values to the simulation that were gracefully degraded by either interpolating across missing values or repeating the last good value. (B) Reading Accuracy vs. Snellen Chart Acuity. With an R2-value also close to zero, a relationship also was not found between performance on the simulated vision task and the visual acuity of normal sight as measured on the Snellen chart task, although a trend was seen where better natural acuity lead to slightly better performance on the task. Again, removing the datum at 76% tracking as an outlier does not affect the outcome. (C) Reading Accuracy vs. Tracking Noise. With an R2-value again close to zero, a relationship was not found between performance on the simulated vision task and noise in the tracking signal. When the same three comparisons were performed for reading speed instead of accuracy, the R2-values were slightly higher, with 0.02, 0.04, and 0.06, respectively, but still close to zero. When performed for reading acuity, they were to 0.01, 0.04, and 0.01, respectively. The combination of these analyses suggest that we did not have instrumentation problems that would have led to unintended bias in the primary results.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Acland G. M., Aguirre G. D., Ray J., Zhang Q., Aleman T. S., Cideciyan A. V., et al. . (2001). Gene therapy restores vision in a canine model of childhood blindness. Nat. Genet. 28, 92–95. 10.1038/ng0501-92 - DOI - PubMed
    1. Anderson S. J., Mullen K. T., Hess R. F. (1991). Human peripheral spatial resolution for achromatic and chromatic stimuli: limits imposed by optical and retinal factors. J. Physiol. 442, 47–64. - PMC - PubMed
    1. Beltran W. A., Cideciyan A. V., Lewin A. S., Iwabe S., Khanna H., Sumaroka A., et al. . (2012). Gene therapy rescues photoreceptor blindness in dogs and paves the way for treating human X-linked retinitis pigmentosa. Proc. Natl. Acad. Sci. U.S.A. 109, 2132–2137. 10.1073/pnas.1118847109 - DOI - PMC - PubMed
    1. Bourkiza B., Vurro M., Jeffries A., Pezaris J. S. (2013). Visual acuity of simulated thalamic visual prostheses in normally sighted humans. PLoS ONE 8:e73592. 10.1371/journal.pone.0073592 - DOI - PMC - PubMed
    1. Brindley G. S., Lewin W. S. (1968). The sensations produced by electrical stimulation of the visual cortex. J. Physiol. 196, 479–493. - PMC - PubMed

LinkOut - more resources

-