Impact of simulated micro-scotomas on reading performance in central and peripheral retina

Abstract

Observers with central field loss typically fixate within a non-foveal region called the preferred retinal locus, which can include localized sensitivity losses, or micro-scotomas (Krishnan and Bedell, 2018). In this study, we simulated micro-scotomas at the fovea and in the peripheral retina to assess their impact on reading speed. Ten younger (<36 years old) and 8 older (>50 years old) naïve observers with normal vision monocularly read high and/or low contrast sentences, presented at or above the critical print size for young observers at the fovea and at 5 and 10 deg in the inferior visual field. Reading material comprised MNREAD sentences and sentences taken from novels that were presented in rapid serial visual presentation (RSVP) format. Randomly distributed 13 × 13 arc min blocks corresponding to 0 – 78% of the text area (corresponding to ~0 – 17 micro-scotomas/deg²) were set to the background luminance to simulate micro-scotomas. A staircase algorithm estimated maximum reading speed from the threshold exposure duration for each combination of retinal eccentricity, contrast and micro-scotoma density in both age groups. Log₁₀(RSVP reading speed) decreased significantly with simulated micro-scotoma density and eccentricity. Across conditions, reading speed was slower with low- compared to high-contrast text and was faster in younger than older normal observers. For a given eccentricity and contrast, a higher density of random element losses maximally affected older observers with normal vision. These outcomes may explain some of the reading deficits observed in older observers with central field loss.

1. Introduction

Dr. Steinbach began his research career by studying the inter-relationships between pursuit eye movements and visual perception. Subsequently, he investigated the question of how eye position is signaled internally, first in normal observers and individuals with strabismus and later in people who had undergone unilateral enucleation. His work in this area led to a serious reconsideration by the vision-science community of how eye-muscle proprioception contributes to the perception of visual direction. About 10 years ago, Dr. Steinbach’s research interests expanded to address the visual and oculomotor characteristics of individuals who lose central vision as the result of age-related macular degeneration. This is one of the many areas in which Dr. Steinbach’s and our research interests overlap. Indeed, we were fortunate to collaborate with Dr. Steinbach and his colleagues on a research project to assess the repeatability of fixation measures in individuals with central-field loss (Bedell et al., 2015). Although the participants in the study reported below had normal vision, the experimental protocol was designed to simulate the presence of localized regions of insensitivity that recently were found to exist within the non-foveal preferred area of fixation of people with macular disease (Krishnan and Bedell, 2018). The outcome measure in our study is reading speed, which Dr. Steinbach and his collaborators also addressed in some of their recent publications (Tarita-Nistor et al., 2018, 2014, 2013a, 2013b). We are extremely pleased to contribute our work to this special issue to honor the extensive and illustrious research career of Dr. Steinbach.

Reading is one of many day-to-day activities that is both visual and cognitive. Vision and its role in reading have been investigated for more than a century (e.g., Huey, 1908). In the past few decades, the task of reading in altered visual conditions, as in observers with real (McMahon et al., 1991) or simulated (Fine and Rubin, 1999) vision loss or with involuntary nystagmus eye movements (Thomas et al., 2011; Woo and Bedell, 2006), has received increased attention. Legge and his colleagues (Legge, Rubin, Pelli, & Schleske, 1985) were one of the first groups to demonstrate a subnormal (~5× slower) reading speed in observers with bilateral central vision loss (CVL). Conditions like age-related macular degeneration (AMD) and Stargardt disease (STGD) are common causes of CVL and visual impairment, which is further compounded by an associated loss in contrast sensitivity (Kleiner, Enger, & Fine, 1988; Ortiz, Jiménez, Pérez-Ocón, & Castro, 2010). Many observers with CVL compensate for their loss of central vision by choosing an eccentric viewing locus or, sometimes, multiple eccentric loci for performing dayto-day activities including reading and facial recognition (Macedo et al., 2011; Sullivan et al., 2008; Timberlake et al., 1986). Although designated in the literature as the “preferred retinal locus” (PRL; Crossland et al., 2011), the area used for fixation by individuals with central field loss typically is larger than that used by individuals with normal vision (Tarita-Nistor et al., 2008; Timberlake et al., 2005; White and Bedell, 1990; Whittaker et al., 1988).

1.1. Rapid Serial Visual Presentation (RSVP) Reading

Altered programming and execution of saccadic eye movements (Frost, 1976), fixation instability (Falkenberg et al., 2007; Whittaker et al., 1988) and extensive crowding (Levi, 2008) are some of the factors that impact performance when individuals are required to read text using a peripheral (non-foveal) retinal location. During RSVP, either the individual letters of a word or the words of a sentence are sequentially presented one at a time, at a given location to reduce the role of saccadic eye movements. RSVP reading has been explored for more than half a century (Forster, 1970; Sperling, 1960), and has been shown to be faster than conventional reading; by a factor of ~1.4× - 2.1× in normally sighted observers (Rubin & Turano, 1994; Yu, Cheung, Legge, & Chung, 2007) and by a factor of ~1.5× in observers with CVL (Rubin & Turano, 1994). Although the RSVP paradigm successfully minimizes the need for saccades, observers with CVL still require longer word durations to read than non-CVL observers (Rubin & Turano, 1994). Rubin & Turano suggested that the peripheral retina includes a severe bottleneck that limits the ability to decode patterns, such as those required in reading.

1.2. Simulated Scotomas and Visual Performance in CVL

Several studies have employed pixelized simulations to better understand the characteristics of visual impairment (Legge et al., 1985; Seiple et al., 1995) or to define the capabilities of prosthetic vision (Cha et al., 1992; Dagnelie et al., 2006; Fu et al., 2006; Sommerhalder et al., 2003). In a seminal study of low vision observers with various etiologies, matrix sampling (by punching holes in a regular array on a plastic transparency and placing it over TV monitor display) was used to limit the amount of available information (Legge et al., 1985). The authors presented drifting text with a wide range (0.5–24°) of letter sizes and reported that the minimal sampling density required for optimal reading (i.e., the critical sampling density) was typically less in observers with low vision, compared to observers with normal vision. In general, the critical sampling density increased with increasing letter size. In a study that investigated the impact of sampling on letter identification, Seiple, Holopigian, Szlyk, & Greenstein (1995) blanked randomly selected pixels in a rectangular area (20 × 27 arc min), within which single letters were presented, to mimic the sampling-element loss that can occur in retinal diseases like retinitis pigmentosa. The luminance of the blanked pixels was set equal to the surround and, interestingly, even high levels (~80%) of pixel blanking didn’t affect performance. Geller, Paul, & Green (1992) reported a similar finding when observers were asked to identify the orientation of high contrast square-wave gratings at or below the Nyquist frequency limit with different levels of stimulus degradation, produced by removing varying percentages of the array elements. The above-mentioned studies assessed the impact of stimulus degradation only on foveal performance for letter or grating identification. Levi, Sharma, & Klein (1997) investigated the number of stimulus samples needed for pattern identification in both foveal and peripheral vision. Of the 17 Gabor features that formed an E shape, about 40–50% were required at the fovea to achieve a threshold level of correct performance, whereas the number of samples required to reach the identification threshold at 5 and 10 deg in the inferior field increased to ~70%.

Walsh & Liu (2014) assessed the impact of both sharp- and gradual-edged scotomas on a visual search task. The sharp-edged scotoma resulted in a more consistent use of a single peripheral location (called the preferred retinal locus,) during search than the gradual-edged scotoma. The results suggest that “scotoma visibility,” which is reduced when the scotoma has gradual edges, might play a role in adapting successfully to CVL (Fletcher et al., 2012; Goodrich, 1977). Liu & White (2010) used random local degradation of texture patterns to assess the impact of this manipulation on texture discrimination in young and elderly observers with normal vision and in patients with early AMD. They reported a reduction in performance and reduced tolerance to stimulus degradation with early AMD. Lastly, Winther & Frisen (2015) used segmented digits (size ~20/200) that were formed by 0.5-arc-min-diameter bright dots to assess macular sensitivity deficits in observers with CVL secondary to AMD. The median number of bright dots per character stroke that was required by groups of normal, dry AMD and wet AMD observers to read digits presented for 150 ms was 3, 5 and 30, respectively. These results are consistent with the idea that localized sensitivity losses at the PRL of observers with CVL would necessitate a greater number of samples per letter stroke to render legible test digits.

Only a few studies employed structural imaging to specifically probe the PRL region of patients with CVL in detail. Nevertheless, structural changes such as a loss of junctional integrity between photoreceptor inner and outer segments, thinning of the photoreceptor layer, and thickening of the retinal pigment epithelium have been reported in association with drusen and other retinal changes (Acton et al., 2012; Fleckenstein et al., 2008; Landa et al., 2011; Rogala et al., 2015; Schuman et al., 2009). Such structural changes have been associated with micro-perimetric visual field defects in cohorts of observers with early AMD (Acton et al., 2012), dry and wet AMD (Landa et al., 2011) and, more recently, individuals with Stargardt disease (STGD, Verdina et al., 2017). Recently, Denniss et al. (2017) found that visual sensitivity typically is impaired near the PRL in individuals with AMD. In a recent study from our group using supra-threshold screening with a micro-perimeter, we found that micro-scotomas exist within the PRL (defined as the retinal region that includes either 68 or 95% of the recorded fixation loci) in nearly 80% of a sample of observers with acquired CVL secondary to conditions like AMD, STGD, and Plaquenil maculopathy (Krishnan and Bedell, 2018). The prevalence of micro-scotomas was similar in the observers with different etiologies of CVL. Because of their relatively small size and the propensity for small retinal scotomas to fill in (Schuchard, 1993; Zur and Ullman, 2003), micro-scotomas at the PRL are likely to remain imperceptible and, hence, it might be difficult for patients with CVL to adapt to their presence.

1.3. Motivation and Predictions

The demonstration that micro-scotomas occur within the word-fixation PRL of many observers with CVL (Krishnan and Bedell, 2018) leads to the question of whether and how these micro-scotomas impact visual performance. Although high densities of simulated micro-scotomas are required to degrade foveal letter recognition or orientation discrimination (Geller et al., 1992; Seiple et al., 1995), an outcome measure such as reading speed might be disrupted more readily by micro-scotomas, especially for text presented in the retinal periphery. The current study therefore evaluated the impact of simulated micro-scotomas on the foveal and non-foveal reading speeds of normal observers. By simulating micro-scotomas with and without an associated contrast sensitivity loss while assessing RSVP reading speed in both younger and older observers, we sought to better anticipate and explain the functional reading loss in patients with CVL. Because the areas of missing information produced by random pixel deletion are tiny and scattered across a region of the visual field we expected their influence to be more akin to a gradual-than a sharp-edged scotoma. For this reason, and because we varied the locations of the micro-scotomas (to simulate the influence of fixational eye movements), we did not expect observers with normal vision to adapt to this manipulation during the experiment. Further, as the simulated micro-scotomas would be expected to be less perceptible for words of low compared to high contrast, we anticipated that the impact of random-element deletion would be more severe when observers attempted to read low- compared to high-contrast text.

2. Methods

2.1. Observers

Eighteen naive observers (10 younger, <36 years, age range: 24–36 years; 8 older, >50 years, age range: 55–73 years) with best-corrected visual acuity of 20/20 or better (and no self-reported ocular comorbidities) were recruited from among the faculty, staff and students of the University of Houston, College of Optometry. The study protocol was approved by the committee for the protection of human subjects at the University of Houston and all observers provided written informed consent before participating. Observers were compensated in part for their time and participation. Each observer’s preferred eye (determined as the eye that observers didn’t close when asked to shut one eye) was chosen for testing and the other eye was occluded with a black opaque patch. Six of the 8 older observers wore their habitual correction (progressive lenses or trifocal contact lenses), which was appropriate for the two testing distances of 35 and 57 cm used for reading in this study. The other 2 older observers read with a near correction that was appropriate for 35 cm and rejected a reduced reading addition for 57 cm. Testing was done under normal room illumination with natural undilated pupils. A chin and a forehead rest were used to stabilize the observer’s head. The reading material used in this study included 60-character MNREAD (Mansfield et al., 1994) sentences (provided by Drs. Steven Mansfield and Gordon Legge) and 53 ± 8 character sentences, taken from novels (provided by Dr. Susana Chung, c.f., Chung, Mansfield, & Legge, 1998). Words were presented in lower case Courier font (Tarita-Nistor et al., 2013b), except for the first letter of the word that began each sentence and the first letter of proper nouns.

2.2. Apparatus

A 21” flat Sony Trinitron cathode ray tube display (model GDM-F520, resolution of 1600 × 1200 pixels) was used to present the test stimuli at a vertical refresh rate of 85 Hz. The test stimuli were designed using the Psychophysics Toolbox-3 (Brainard, 1997; Kleiner et al., 2007) that interfaces between MATLAB R2014a (Mathworks, Natick, MA, USA) and the computer hardware used. The test area comprised 1280 × 1024 pixels and subtended 36° × 29° at the 57 cm viewing distance (1 pixel subtended ~1.7 arc min). The luminance of the dark letters and the uniform white background (L_BG ≈ 113 cd/m²) was assessed using a Minolta LS-100 luminance meter. Testing using letters with high (> −90%) and low Weber contrast (≈ −10%) was performed in 2 different sessions. All 8 older observers and 5 of the younger observers participated in both high and low contrast testing. Three young observers participated only in the high-contrast condition, while 2 young observers completed only the low contrast testing. During the study, the examiner sat next to the subject and monitored his or her fixation visually. Trials that were observed to include a vertical saccade were discarded (estimated to be <10%) and re-run immediately.

2.3. Pilot Testing for Reading Acuity

In a pilot study involving 5 younger naïve observers with normal vision (ages: 25–31 years), the word reading acuity was estimated at the fovea (E0) and at 5 and 10 deg (E5 and E10) in the inferior visual field. Observers fixated on a 1-deg cross (turned off during foveal testing) and randomly selected words of 5–10 letters sampled from the sentence pool described above were displayed at various sizes (acuity range: 20/800 – 20/20 in 0.1 logMAR steps) to determine the reading acuity, defined as the smallest letter size at which words could be read correctly. Testing was repeated thrice for each observer, using both high and low contrast words at eccentricities E0, E5, and E10. The critical print size was specified as 4 step sizes larger than the reading acuity, i.e., critical print size = mean logMAR reading acuity + 0.4 (Cacho et al., 2010; Chung et al., 1998). In the main experiment, the x-height (height of the lowercase ‘x’, c.f., Mansfield et al., 1994) of the text presented at each eccentricity was equal to the average critical print size as determined in the pilot study, with the exception of high-contrast testing at the fovea where the x-height was larger (x-height = logMAR reading acuity + 0.7). Larger letters were used in the high contrast foveal condition to prevent the individual simulated 13 × 13 arc min micro-scotomas (see below) from obscuring entire letters. Recently, Calabrèse et al. (2016) documented that several reading measures, including the critical print size, change systematically with age in normally sighted observers. Specifically, Calabrese et al. (2016) reported that the critical print for normal observers increases by 0.003 logMAR/year between 23 and 68 years old and by 0.01 logMAR/year for observers between 68 and 81 years old. As noted above, our determination of the critical print size was based on results obtained from younger observers with an average age of 27.9 years old. Therefore, we may have underestimated the critical print size for our older observers, whose average age was 60.5 years old, by approximately 0.1 log units (0.003 × [60.5 – 27.9]). The potential adverse influence of this underestimation of the critical print size on the reading speeds achieved by the older observers in our study is addressed in section 4, below.

2.4. RSVP Testing for Reading Speed - Simulation of Micro-scotomas

At each eccentricity tested (E0, E5 and E10) the words of a randomly selected sentence from the sentence pool were presented sequentially, one at a time. As noted above, when the text was presented at E5 and E10, the experimenter monitored the observers to ensure that fixation was maintained on the fixation cross. Each presented word was centered both horizontally and vertically within a rectangular RSVP window (Figure 1, right), which itself was at the center of the display screen. In different conditions, randomly distributed 13 × 13 arc min blocks (chosen to match the size of the stimuli used previously to map micro-scotomas in patients with CVL by Krishnan and Bedell, 2018) corresponding to 0, 13, 26,…,78% of the RSVP window were set to the background luminance (gray scale = 255) to simulate the influence of localized micro-scotomas (Figure 2). The corresponding values in number of micro-scotomas/deg² were: 2.8, 5.5, 8.3….,16.6 respectively. The step size of 13% (2.77 microscotomas/deg²) was chosen to ensure adequate sampling of the range of micro-scotoma densities from 0 to 75% and is not related to the image-block size of 13 arc min. To simulate the influence of fixational eye movements, the location of all the micro-scotomas was jittered (SDx, SDy = 0.3 deg) in tandem during the presentation of the successive words in each sentence. Both the locations of the simulated micro-scotomas within the RSVP window and the jitter to simulate the effect of fixational eye movements varied from sentence to sentence.

Figure 1.

(Left) A representation of the sequence of words of a sentence presented during the assessment of RSVP reading speed using high-contrast text, at an eccentricity of 5 deg in the inferior visual field (E5). The fixation cross and the vertical attention markers stayed on the screen throughout each trial. (Right) The orientation screen is enlarged to show details. Note that the margin of the RSVP window was not visible during testing (See Figure 2). The padding area (empty white space inside the RSVP window) that surrounds the word in both the horizontal and vertical directions allows for simulated fixational jitter of the micro-scotoma locations.

Figure 2.

A cropped image showing the central portion of the screen during RSVP testing. The example shows two high-contrast words at E5 (x-height = 1.2 logMAR or 1.32 deg) and at a simulated micro-scotoma density of 52% (11.1 micro-scotomas/deg²). Note that the same micro-scotoma density can have a different impact on the visibility of the individual letters and the word depending on the specific letters that comprise the word and the locations of the micro-scotomas with respect to the letter strokes. As seen here, the RSVP window margins were not visible during testing.

2.5. RSVP Window

The RSVP window, excluding a surrounding padding area, was 13 letter characters wide and 2 characters high. The surrounding padding area (±0.9 deg) was included to accommodate the word-to-word jitter that was introduced in the location of the micro-scotomas. The RSVP window size was contingent on the text x-height (window height = 2*(x-height)) and so was different for the 3 eccentricities tested. A fixation cross was present throughout testing, except for testing at E0. Vertical markers (Figure 1, right), which helped the observers to direct their attention to the appropriate location on the display screen for words at eccentricities of E5 and E10, stayed on the screen only during the presentation. The attention markers appeared with onset of the first word and disappeared simultaneously with offset of last word in each sentence. The height of the attention markers was set equal to one x-height and the vertical center-to-center distance between each word and surrounding markers was set to 3 times the x-height. These values were so chosen to minimize any interference between the attention markers and the margins of the RSVP window.

2.6. Reading Speed Assessment

After about 15 minutes of demonstration, training and practice with the RSVP reading task, assessment began either at E5 or E10, chosen randomly across observers. The chosen condition was always followed by testing at the fovea and then at the third, remaining eccentricity. For the observers who participated in testing using both high and low contrast text, the high contrast condition was always done first. An auditory beep signaled the start and the end of each trial and the observers either read the words as they appeared sequentially on the screen or read the whole sentence after all the words had appeared (Figure 1 left). There was no time limit to respond and observers were encouraged frequently to guess and to make corrections if deemed necessary.

Individual sentences were shown just once to each observer. Following the observer’s response on each trial, the whole sentence was displayed on the screen to provide immediate feedback. Presenting the entire sentence on the display screen also helped the examiner to score the observer’s response to each sentence as either pass (read aloud all the words correctly, in any order) or fail (missed one or more words). The examiner recorded the score using either the ‘up’ or ‘down’ arrow key on the computer keyboard, which increased or decreased the exposure duration of each word on the following trial. The different levels of micro-scotoma density were tested in ascending order, starting from 0% to 39%, and to 52% (shown in Figure 2), 65% and 78%, if reading was still possible.

A staircase algorithm estimated the threshold exposure duration for each combination of retinal 282 eccentricity, letter contrast and micro-scotoma density. The exposure duration of each word changed in gross steps of 50% until the occurrence of the first staircase reversal, after which a step size of 25% was used to define 3 more reversals. For each condition tested, the reading speed in words per minute (wpm) was determined for each subject from the threshold exposure duration, as shown in Figure 3.

Figure 3.

Illustration of computation of threshold word exposure duration determined using a staircase algorithm. Two representative scenarios (A and B) are shown. The exposure duration of the first sentence (S1, shown here as 1000 ms) was selected based on the test conditions (e.g., longer for low contrast text and peripheral testing). The exposure duration immediately after the first reversal (shown here using the bigger block arrows) was not considered for threshold computation. The staircase continued until 3 reversals were obtained at a 25% step size (smaller block arrows) and the RSVP reading speed was determined using the formula in the inset, adapted from the expression used to calculate MNREAD reading speed (Mansfield et al., 1994).

2.7. Data Analysis

A linear mixed-effects modeling approach (Bagiella et al., 2000) was used to assess the effect of simulated micro-scotoma density (% deletion) on the log₁₀(RSVP reading speed) using PROC MIXED in SAS 9.2 (SAS Institute, Cary, NC) and allowed for the primarily within-subject nature of the study design while handling missing data. Reading speed is expressed throughout as log₁₀(wpm) to approximately equalize variances across conditions. A random effect for subject with a variance component structure was specified and the residual correlation was modeled specifying an unstructured covariance structure. We also were interested in the effects of contrast, eccentricity, and age on mean reading speed and how these factors affected the relationship between micro-scotoma density and log₁₀(RSVP reading speed). Therefore, fixed effects included within-subject variables micro-scotoma density as a continuous variable, repeated factors contrast (2 levels: high, low) and eccentricity (3 levels: 0, 5, 10 deg) and the between individual factor age (2 levels: younger, older) in addition to interaction terms. We checked the interaction terms, beginning with the full model including highest order terms, and removed non-significant interaction terms to simplify the model, while also aiming to minimize the Akaike information criteria (Akaike, 1974). A residual analysis was used to evaluate the fit of the final model. The parameter estimates for the main effects and interaction terms included in the final model are presented in Table 1 along with p-values based on the t statistic.

Table 1.

Maximum likelihood parameter estimates of log10(RSVP reading speed in wpm) for the linear mixed-effects model. The second column lists the estimate (b) and standard error (SE) for each fixedeffect parameter. The parameter estimates are for comparison with the arbitrarily chosen reference condition of: high contrast text, older age group and E0 (fovea).

Effect	Estimate (SE)	p-value
Fixed Effects: Main	[b (SE)]
Intercept	2.742 (0.07)	<.0001
Contrast: Low	−0.305 (0.06)	<.0001
Age: Younger	0.259 (0.09)	0.0107
Micro-scotoma density	−0.025 (0.001)	<.0001
Eccentricity: E5	−0.43 (0.06)	<.0001
Eccentricity: E10	−0.402 (0.06)	<.0001
Fixed Effects: Interactions
Micro-scotoma density * Eccentricity (E5)	0.002 (0.002)	0.2708
Micro-scotoma density * Eccentricity (E10)	0.014 (0.002)	<.0001
Micro-scotoma density * Age (Younger)	0.004 (0.002)	0.0274
Micro-scotoma density * Contrast (Low)	−0.005 (0.002)	0.0059
Contrast (Low) * Eccentricity (E5)	0.327 (0.069)	<.0001
Contrast (Low) * Eccentricity (E10)	0.325 (0.06)	<.0001

3. Results

In the preliminary experiment, the average critical print sizes for E0, E5 and E10 were 0.4, 1.2 and 1.7 logMAR (0.21, 1.32 and 4.18 deg) respectively, for high contrast words and 0.7, 1.2 and 1.7 logMAR (0.42, 1.32 and 4.18 deg) for low contrast words. For RSVP testing with high contrast words and no (0%) simulated micro-scotomas, the average reading speeds at the fovea, in log₁₀(wpm) ±SD and actual (wpm) were: 2.87±0.12 (743.2 wpm) and 2.77±0.16 (584.1 wpm) for the younger and older observers, respectively. The corresponding values for testing with low-contrast words were: 2.69±0.08 312 (492.4 wpm) and 2.49±0.13 (310.8 wpm).

As shown in Figure 4, log₁₀(RSVP reading speed) decreases significantly with the simulated micro-scotoma density (b = −0.025, p<0.0001, also see Table 1). In the absence of simulated micro-scotomas foveal reading speed was faster than at eccentricities of 5 (for E5, b = −0.43, p<0.0001) and 10 deg (for E10, b = −0.402, p<0.0001). However, with an increase in the percentage of simulated micro-scotomas, log₁₀(reading speed) falls off less rapidly in the E10 condition than in the E0 and E5 conditions. Indeed, the slopes of the linear fits (across both contrasts and age groups) are significantly shallower at an eccentricity of E10 vs. E0 (_tdf-=89.4 = 8.3, p<0.0001) and at E10 vs. E5 (t_df=74.4 = 6.8, p<0.001). The fitted slopes for E0 vs. E5 are not significantly different (t_df=55.25 = 1.19, p=0.24, also refer to Figure 5C).

Figure 4.

Log₁₀(RSVP reading speeds, in wpm) as a function of % simulated micro-scotoma density for high- (top row) and low-contrast text (bottom row) in younger (left column) and older (right column) observers. A 13% micro-scotoma density corresponds to 2.8 micro-scotomas per square degree of visual angle. Note that testing for >39% simulated micro-scotoma density for low- and >52% for high-contrast words was possible only at an eccentricity of 10 deg (E10). Not all observers could be tested for the highest micro-scotoma densities (please refer to the supplementary data table) and hence the reading speeds for highest micro-scotoma densities at E5 (HC-Older) and E0 (LC-Older) are more variable and less reliable. The shaded region around each plotted function shows the 95% confidence interval for each eccentricity.

Figure 5.

Figure 5A Plots illustrating the interaction of age group and simulated micro-scotoma (MS) density on log₁₀(reading speed).

Figure 5B Plots illustrating the interaction of contrast and simulated micro-scotoma (MS) density on log₁₀(reading speed).

Figure 5C Plots illustrating the interaction of eccentricity and micro-scotoma (MS) density on log₁₀(reading speed)

(A-C). The linear fits from the mixed-effects model, with micro-scotoma density on the X-axis and the predicted log₁₀ (reading speed) on the Y-axis. The shaded region represents the 95% confidence limits for each fitted line. Note that Figure 5C is the model fit to the plot of raw data shown in Figure 4

The expected reading speed was slower (10^0.305 ≈ 2x) for the low- compared to the high-contrast text and was faster (10^0.259 ≈ 1.8x) in the younger compared to the older observers (see Table 1, also Figure 5A and Figure 5C). The intercept value of 2.74 (10^2.74 ≈ 552 wpm) represents the model-predicted mean log₁₀(RSVP reading speed for the foveal testing of older observers with high contrast words [reference condition, see Table 1]) at 0% simulated micro-scotoma density. The 2-way interactions that were significant (p<0.05) are micro-scotoma density x eccentricity, micro-scotoma density x contrast, and contrast x eccentricity (from type-3 tests of fixed effects, not shown here). The values listed in Table 1 confirm that the relationship between micro-scotoma density and reading speed is significantly modified by eccentricity (E10 vs. E0, b=0.014, p<0.0001), age (younger vs. older b=0.004, p=0.0274), and contrast (low vs. high, −0.005, p=0.0059).

4. Discussion

Our study demonstrates that simulated micro-scotomas exert a negative impact on RSVP reading performance both in the central and peripheral retina. Although this is not surprising, it is worth noting that the impact was persistent even after scaling the word size to be equal to the critical print size at each eccentricity. The mean critical print size from clinical studies that tested hundreds of observers with CVL ranges from 1.2 – 1.43 logMAR (Cacho et al., 2010; Fletcher, Schuchard, & Gale, 1999; Legge et al., 1985). Thus, the x-height of the words used in our study at 5 and 10 deg in the inferior visual field is either equal to or larger the mean values of critical print size for CVL observers. Impaired reading performance for low contrast words is not a novel result (Legge & Kersten (1987), but it may help to explain the interaction of scattered sensitivity losses (from micro-scotomas) with impaired contrast sensitivity, such as those seen in some individuals with CVL secondary to macular degeneration (Rubin & Legge, 1989). Taken together, our findings shed light on the reading deficits in individuals with CVL. The greater the number of micro-scotomas, and the more contrast sensitivity is impaired, the slower should be the reading speed. Because the RSVP mode of text presentation minimizes the need for reading eye movements, the impact of micro-scotomas on real-world page reading can potentially be more detrimental than shown by our results.

In the absence of simulated micro-scotomas, the average reading speeds of our observers at eccentricities of 5 and 10 deg are very similar (Figure 4). Latham and Whitaker (1996) reported similar rates of word recognition at eccentricities between 0 and 10 deg when the letter size is scaled appropriately. However, in agreement with several subsequent studies (Bernard et al., 2013; Chung, 2004, 2002; Chung et al., 1998), the reading speed for meaningful text was notably faster in the fovea than peripherally. Nevertheless, for some observers very similar results were obtained at eccentricities of 5 and 10 deg for text that is equal to or larger than the critical print size (Chung, 2004, 2002; Chung et al., 1998). The data in our Figure 4, indicate that, as the percentage of micro-scotomas increases, reading speed decreases more gradually at an eccentricity of 10 deg compared to the fovea and 5 deg. This outcome appears to be both counterintuitive and at odds with the report by Legge et al. (1985) that the critical sampling density becomes larger with increasing (i.e., more peripheral) letter size. However, in contrast to the simulated micro-scotomas in our study, which had a fixed size of 13 × 13 min arc, Legge et al. (1985) produced sampling using masks that were scaled in proportion to the size of the presented letters. We propose that the relative size of the letters compared to the simulated micro-scotomas used in our study can account for different rates of fall-off of reading speed that we observed at eccentricities of 0, 5 and 10 deg. Specifically, even at relatively low simulated micro-scotoma densities a single 13 × 13 min arc element deletion is sufficient to obscure one full quarter of a 0.7 logMAR foveal letter, potentially rendering it unrecognizable. In contrast, a substantially higher density of 13 × 13 min arc simulated micro-scotomas is necessary to obscure a full quarter of the 1.7 logMAR letters presented at an eccentricity of 10 deg. As the sizes of the actual micro-scotomas that occur in individuals with central field loss are likely to vary from patient to patient, application of the relative rates of fall off in reading speed at the different eccentricities we tested should be treated with considerable caution.

Of the other significant interactions that we identified, two stand out and warrant comment. In addition to an overall faster reading speed in the younger- compared to the older-subject group (Table 1, p < 0.0107), the reading speed of the younger observers drops slightly but significantly more slowly as the density of simulated micro-scotomas increased (Table 1, b= 0.004, p=0.0274). As noted in section 2.3, above, the critical print size for text presented at the fovea increases with age (Calabrese et al., 2016), although it is not known whether the critical print size changes with age for non-foveal reading. In any event, our determination of the critical print size using only younger observers may have disadvantaged the older observers in our study. However, Calabrese et al. (2016) reported also that the average maximum reading speed decreases with age (see also Brussee et al., 2017) from approximately 201 wpm at age 27.9 (the average age of the younger observers who determined the critical print size) to 188 wpm at age 60.5 (the average age of our older observers). As reported above in section 3, the average reading speeds of our younger and older observers for high contrast text presented at the fovea with no simulated micro-scotomas were 743 and 584 wpm, respectively. However, the text presented in this condition was 0.7 log units larger than the estimated value of reading acuity (see section 2.3), which makes it highly likely that the critical print size was exceeded both for the younger and older observers. We therefore conclude that at least some portion of the differences in reading speed between the younger and older groups of observers in this study should be attributed to age and not to a difference in the critical print size. Both the slower overall reading speed as well as the more deleterious impact of random localized sensitivity losses in the older compared to younger observers might result from causes like altered high-level visual processing (such as visual memory) in the older-age cohort, and reduced contrast sensitivity in older individuals.

The significant interaction between micro-scotoma density and contrast is noteworthy. With other factors being the same, reading speed falls more quickly with the density of simulated micro-scotomas (b= −0.005, p=0.0059) for low than high contrast text. One way to interpret this interaction is that, by setting small regions of the text to the background luminance, the introduction of simulated micro-scotomas can effectively reduce the local contrast. Rubin & Legge (1989) reported a rapid reduction in reading speed for contrasts below the critical contrast, the text contrast at which reading speed falls to one half the maximum value. Because the contrast “reserve” is smaller for low- compared to high-contrast text, a reduction of the effective contrast that results from simulated micro-scotomas would be expected to exert a greater influence on reading speed for low-contrast words.

Although we simulated an influence of fixational eye movements on successively presented words by jittering the locations of simulated micro-scotomas en bloc both horizontally and vertically, the fixational eye movements made by observers with CVL may differ quantitatively from this simulation. For example, Kumar & Chung (2014) reported that the median amplitude of slow fixational drifts in 16 observers with macular disease (mean age ~75 years) was ~14 arc min (0.23°) and the median amplitude of micro-saccades was ~53 arc min (0.88°), approximately 2 and 3.5 times larger than the corresponding fixational eye movements of age-matched observers with normal vision. Thus, our simulation of fixation jitter (SDx, SDy = 0.3 deg) may, to some extent, under-represent the distribution of the fixation eye movements in patients with CVL.

Scherlen, Bernard, Calabrese, & Castet (2008) assessed page-mode reading in 7 observers with normal vision in the presence of a simulated artificial scotoma (6 or 10 deg in diameter) that was filled with upper-case X characters and concluded that the number of saccades and the mean fixation duration are the primary factors that correlate with reading speed. The authors suggested that when the visual encoding of text becomes more difficult, more saccades are required within a given region of text for word identification to occur. This interpretation is consistent with work by Deubel, Schneider, & Bridgeman (2002), who reported that when visual encoding is degraded the efficiency of trans-saccadic integration of information is reduced. Put together, the degradation of visual input potentially necessitates an increase in fixation duration (Loftus et al., 1992), thereby slowing the reading speed. Extending these observations to our study, the RSVP paradigm nullifies the need for saccades across the words of a sentence and the need for encoding the relative spatial locations of text across eye movements. However, the degradation of the text that is produced by simulated micro-scotomas still may interfere with the integration of information at low, and perhaps also at high levels of visual processing.

Rayner, Pollatsek, & Schotter (2012) compared template and feature models of word recognition during reading. In template models, visual input is compared to the memory representations of various objects (templates) and the best matching template is perceived as the object. On the other hand, feature models posit that recognition occurs based on the individual elements that constitute the object (like the combination of strokes that form a specific letter and/or the letters that comprise a word). Irrespective of which (if either) type of model is correct, the degradation produced by simulated micro-scotomas can impact recognition and reading performance, either because missing areas of the stimulus make template matching difficult or because the individual features that comprise letters and words are distorted or degraded. Post-receptoral under-sampling in normal peripheral retina has been well explored and documented (Anderson and Hess, 1990; Campbell and Green, 1965). Demirel et al. (2012) proposed a neural sampling model to account for the detection and identification of ‘vanishing optotypes’ (high-pass-filtered stimuli that become undetectable when resolution fails) in central and peripheral vision. In this model, the threshold for letter recognition is determined by the differential activation of a minimum number of receptive fields, which depends both on the size of the presented letter its and contrast. Clearly, introducing random local deletions within the target letter also would be expected to reduce the number of differentially activated receptive fields and increase the letter size needed to achieve recognition.

McMahon, Hansen & Viana (1991) described reading as a complex process requiring visual resolution, stable images, accurate saccades, word encoding, lexical assessment, and short and long-term memory. The last aspect of this process perhaps needs mention, as in our study the observers verbalized words either as they appeared on the computer monitor, following the end of trial, or a mixture of both. The degraded visual stimuli that we used in our study could be argued to have either less or more impact during page-mode, compared to RSVP, reading. The argument in favor of less impact is that during page reading, one may have a preview benefit from text removed from the PRL that does not suffer from localized deletions and perhaps a reduced need to store the visual input in memory. On the other hand, the text degraded by micro-scotomas can potentially have an adverse impact on eye movements and the fact that page mode reading requires efficient eye movements argues for a greater impact of simulated micro-scotomas on page-mode reading.

4.2. Limitations

The simulated micro-scotomas that we employed in this study assume a random distribution of localized sensitivity losses, spanning a considerable portion of the central or the peripheral retina. It may be that such random losses don’t occur in observers with CVL, whose sensitivity losses (even outside the large clinically documented central scotoma) may be more clustered and restricted only to specific regions near the PRL (Denniss et al., 2017). The model we used deleted information in blocks of a fixed size (13×13’) from the region where text was presented. This block size was chosen to match the stimulus size used in our previous demonstration of micro-scotomas within the PRL of observers with CVL (Krishnan and Bedell, 2018). The impact of the simulated micro-scotoma size on reading performance remains to be explored. As a logical extension of the current study, it would be interesting to use maps of the micro-scotomas determined over an extended retinal region surrounding the PRL in observers with CVL to simulate vision loss and assess reading speed in observers with normal vision.

As discussed in section 2.3 as well as above in the Discussion, by using a critical print size that was estimated from the results of younger observers, we might have made the reading task more difficult for our older observers. Although the poorer reading performance of our older observers cannot be attributed wholly to an increase in the critical print size in these individuals (see above), the observed differences in reading performance between the younger and older cohorts should be interpreted with some caution.

Peripheral aberrations, both internal (optics of the eye) and external (from the observers’ optical corrections), could have played a role in our observers’ reading performance at eccentricities of 5 and 10 deg. However, our testing paradigm was intended to mimic real-world viewing in observers with central vision loss, which typically employs the observers’ prescribed refractive correction.

5. Conclusions

RSVP reading speed in both the central and peripheral retina is influenced by the density of random element losses, age, word eccentricity, and text contrast. For a given eccentricity and contrast, higher densities of random element loss maximally affect older observers with normal vision. This may partly explain the subnormal reading performance in older observers with bilateral CVL, who use a retinal location that can include local areas with sub-clinical local sensitivity losses. The results presented here indicate that scattered sensitivity losses near the PRL, in conjunction with co-existing changes in contrast sensitivity, should exert a negative impact on reading speed.

Research Highlights.

• Simulated random element loss impairs reading performance both in central and peripheral vision

• The impact of simulated element loss increases with density and age and decreases with retinal eccentricity and text contrast

• Reading deficits in patients with central vision loss can be partly explained by scattered micro-scotomas

Abbreviations:

RSVP

Rapid Serial Visual Presentation

MNREAD

Minnesota Read

CVL

Central Vision Loss

AMD

Age-related Macular Degeneration

STGD

Stargardt Disease

PRL

Preferred Retinal Locus

logMAR

Logarithm(base10) of the Minimum Angle of Resolution

WPM

Words Per Minute