Perceptual Learning in Children With Infantile Nystagmus: Effects on Visual Performance

Abstract

Purpose: To evaluate whether computerized training with a crowded or uncrowded letter-discrimination task reduces visual impairment (VI) in 6- to 11-year-old children with infantile nystagmus (IN) who suffer from increased foveal crowding, reduced visual acuity, and reduced stereopsis.

Methods: Thirty-six children with IN were included. Eighteen had idiopathic IN and 18 had oculocutaneous albinism. These children were divided in two training groups matched on age and diagnosis: a crowded training group (n = 18) and an uncrowded training group (n = 18). Training occurred two times per week during 5 weeks (3500 trials per training). Eleven age-matched children with normal vision were included to assess baseline differences in task performance and test–retest learning. Main outcome measures were task-specific performance, distance and near visual acuity (DVA and NVA), intensity and extent of (foveal) crowding at 5 m and 40 cm, and stereopsis.

Results: Training resulted in task-specific improvements. Both training groups also showed uncrowded and crowded DVA improvements (0.10 ± 0.02 and 0.11 ± 0.02 logMAR) and improved stereopsis (670 ± 249″). Crowded NVA improved only in the crowded training group (0.15 ± 0.02 logMAR), which was also the only group showing a reduction in near crowding intensity (0.08 ± 0.03 logMAR). Effects were not due to test–retest learning.

Conclusions: Perceptual learning with or without distractors reduces the extent of crowding and improves visual acuity in children with IN. Training with distractors improves near vision more than training with single optotypes. Perceptual learning also transfers to DVA and NVA under uncrowded and crowded conditions and even stereopsis. Learning curves indicated that improvements may be larger after longer training.

Children are categorized as visually impaired (VI) based on their visual acuity and/or visual field restriction. But in childhood, mid- to high-level processes, such as visual search and motion processing, may be impaired too.^1–3 Another debilitating factor is crowding. Crowding refers to a decreased ability to identify objects in clutter.^4,5 Recent studies indicate that perceptual learning can enhance visual performance.^6,7 Perceptual learning is defined as “any relatively permanent and consistent change in the perception of a stimulus array, after practice or experience with this array” (p. 29).⁸ It is considered a manifestation of cortical plasticity, and changes may last for months or even years.^6–9 A prerequisite for effective perceptual learning in children is that they cooperate and have a at least a minimal amount of concentration.¹⁰ Here, we evaluated the effect of computerized training on visual performance of children with infantile nystagmus (IN) occurring in association with albinism or occurring solitarily without an afferent visual defect (a condition called idiopathic IN). 

Albinism and idiopathic IN are two of the most prevalent diagnoses associated with VI in children: 8% of VI children have albinism and 6.6% have idiopathic IN.¹¹ Infantile nystagmus is present at birth or develops in the first 3 to 6 months.¹² Idiopathic IN is characterized by a high-frequency horizontal nystagmus, mildly reduced visual acuity (mean acuity 0.3 logMAR), strabismus (7.8–44%), and relatively good stereopsis.^13,14 In adults with albinism, visual acuity varies from 1.3 to 0 logMAR (mean acuity 0.6 logMAR), virtually all have nystagmus,¹³ there is high prevalence of strabismus, and stereopsis is often absent.^14–16 

Visual acuity of children with IN is often more severely degraded by the presence of distractors than it is in children with normal vision.^17,18 Accumulating evidence indicates that IN is associated with stronger visual crowding and contour interaction.^4,18–20 Factors that contribute to increased crowding and contour interaction in subjects with IN are the presence of amblyopia,¹⁹ fixational instability,^20,21 and nystagmus frequency and amplitude.¹⁸ Crowding is a ubiquitous, well-documented phenomenon in peripheral vision.⁵ Crowding can also occur in foveal vision,^22–24 but in the fovea the phenomenon is quite distinct from peripheral crowding. It is mainly attributed to contour interaction (overlap masking²⁵), inaccurate eye movements,²¹ and response competition between flankers and target.^26,27 A recent study on crowding in 3- to 15-year-old children with normal vision demonstrated that vision in children younger than 7 years is dominated by suppressive interactions between contours.²⁸ Crowding can be measured in two different ways: one is the maximum distance at which a distractor causes degraded perception of the target (the extent), and the second is the amount of loss in visual acuity when the interaction is at its maximum (the intensity).⁴ The crowding extent is the minimum distance needed between a threshold-size target and flankers to allow target recognition.²⁹ The crowding intensity can be captured by the difference between crowded and uncrowded acuity (in logMAR). In this study, we looked at both of these aspects of crowding. 

Perceptual learning refers to performance changes brought about through practice or experience that improves a subject's ability to respond to the environment.³⁰ This form of learning can be explained by changes in low levels of visual processing such as changes in activity in primary visual cortex in correlation with perceptual learning.³¹ In addition, there is evidence that high-level cortical processes are involved in perceptual learning.³² Perceptual learning can be characterized by fast and slow learning.³³ Fast learning typically takes place over the first one to four training sessions (typically a few hundred trials).³³ Slow learning takes place over days.³⁴ The first phase may reflect the setting up of a task-specific routine for solving a visual problem, while the second phase may indicate long-term structural modification of basic perceptual modules.³⁵ A well-documented restriction of perceptual learning is that training effects are often very task specific, which can be seen as a manifestation of plasticity in low-level sensory cortical processes.³⁶ This task specificity may explain why perceptual learning is not yet widely applied in clinical settings.³⁶ 

A possible remedy to promote more generalized learning is to combine multiple perceptual learning approaches: the engagement of attention, the use of reinforcement techniques, and the presentation of stimuli in multiple dimensions/orientations.³⁶ Based on these principles and the known oculomotor deficit in children with IN, we developed a computerized perceptual learning paradigm that integrates two approaches: learning to identify crowded letters, which improves crowded visual acuity and reduces crowding,^6,7,37 and visuomotor training involving the execution of goal-directed saccades under time constraints. We used 10 training sessions with a total of 3.3 hours of perceptual learning. Two previous studies that adopted a short training period (3.4 hours) in adults with amblyopia reported mean acuity improvements of 0.085 logMAR³⁸ and 0.13 logMAR.³⁹ There is also evidence that saccade latencies can be shortened by practice.⁴⁰ By training children to execute goal-directed saccades, they could learn to disengage attention from the central stimulus and make an accurate and timely saccade to the target. In this way, children might improve their performance on other tasks that rely on eye movements, such as reading. There is indeed evidence that goal-directed saccades are delayed in children with IN even though their visual processing speed remains unaffected.⁴¹ 

A drawback of the existing literature on perceptual learning efforts to ameliorate crowding is that many studies do not compare the effect of crowded and uncrowded training paradigms,⁶ making it unclear what the effect of the crowded intervention is compared to a similar training without crowding. The present study did include an uncrowded training group. The crucial difference between the two training protocols was that the uncrowded training involved identifying an uncrowded optotype in the absence of distractors while the crowded training entailed identifying an optotype under crowded conditions. In addition, we included children with normal vision to assess task performance in a reference group and to assess possible test–retest learning effects. 

Methods

Participants

Our group of children with IN was composed of 36 children, of whom 18 had oculocutaneous albinism and 18 had idiopathic IN. The diagnosis of the children was obtained after a full ophthalmologic investigation at the start of their rehabilitation. We also included 11 children with normal vision. Children with normal vision were not trained because the burden of the training was considered too large for these children given that they had no visual acuity deficits. This control group was included to assess baseline differences in performance on the training task and to evaluate test–retest learning. 

Inclusion criteria for all children were age 6 to 11 years, normal birth weight (>3000 g), birth at term (>36 weeks), no perinatal complications, and normal development. For children with IN, additional criteria were no additional impairments and crowded distance visual acuity (DVA) between 0.2 and 1.3 logMAR. Children with normal vision had to have a crowded DVA of 0.1 logMAR or better. Supplementary Table S1 presents the diagnosis, clinical characteristics, and refraction corrections of participants with IN. Refractive errors of all children with IN were checked no longer than 12 months before the start of the training. Children with prescription glasses always wore them during pre- and postmeasurements and training. 

After explanation of the nature and possible consequences of the study, informed consent was obtained from the parents of all participants. The study was approved by the local ethics committee (CMO Arnhem-Nijmegen, The Netherlands) and conducted according to the principles of the Declaration of Helsinki. 

Ophthalmologic Examination

Children with IN underwent ophthalmologic assessment before and after training. Children with normal vision underwent ophthalmic assessment only at baseline. Distance visual acuity (DVA) was measured mono- and binocularly at 5 m with the Landolt C-test, using a crowded chart version with an interletter spacing of 2.6 arcmin and an uncrowded chart version with an interletter spacing of at least 30 arcmin.^41,42 We chose this chart with tight interletter spacing because of its sensitivity to detect crowding effects.^42–45 Its fixed interletter spacing can be transformed to proportional interletter spacing by using the following formula: interletter spacing in arcmin/letter size in arcmin (10^logMAR×5). At 5 m, this C-test has upper and lower limits of −0.1 and 0.8 logMAR, respectively. Therefore, if children had a DVA less than 0.8 logMAR, measurements were taken at 2.5 m whereas a distance of 8.5 m was used if their DVA exceeded −0.1 logMAR. 

Near visual acuity (NVA) was determined binocularly with the LEA version of the C-test at 40-cm viewing distance. The LEA test consists of the same uncrowded and crowded chart versions as the C-test but presents letters in a larger size range of −0.3 to 1.7 logMAR.¹⁷ Viewing distance was monitored carefully with a ruler during the NVA measurements, and the chart was positioned on a reading stand. Stereopsis was measured with the Titmus Stereo Fly test at 40 cm.^46,47 Lighting conditions were controlled during measurements and ranged from 680 to 760 lux. 

Crowding intensity was calculated from the difference between crowded and uncrowded acuities in logMAR.^17,48 Note that crowding intensity equals the logarithm of the crowding ratio, where the crowding ratio is the ratio of crowded acuity to uncrowded acuity in degrees of visual angle. For example, a crowding intensity of 0.3 logMAR indicates that the visual acuity loss due to distractors is 3 logMAR lines or half the uncrowded letter acuity. 

Computer-Controlled Letter-Discrimination Tasks

Subjects also performed two computer-controlled letter-discrimination tasks on a 32-inch liquid crystal display (LCD) to determine their uncrowded VA and crowding extent under a 500-ms time constraint (Fig. 1A). Letters were always high-contrast (99.7% Michelson) black Landolt Cs (0.3 cd/m²) with four possible orientations against a white background (193.8 cd/m²). 

Figure 1

 

Automated letter-discrimination tasks and training tasks. (A) In the single-letter task, the first screen of each trial showed a single Landolt C in the center of the screen. After 500 ms, the C disappeared and a blank response screen appeared until the child pressed a response button. Subsequently, a feedback screen appeared for 200 ms containing a red or green smiley dependent on the correctness of the response. In the crowded letter task, the first screen of each trial showed a patch of seven equally sized Cs of the same orientation. This first patch was presented for 500 ms. The second screen presented an additional patch to either the left or right of the center patch and was shown for 500 ms. The target C was always placed in the center of the second patch and had a different orientation than the surrounding Cs. The third screen was a blank response screen, which was presented until the child pressed a response button. Once the child responded, a feedback screen appeared for 200 ms indicating the correctness of the answer. (B) The uncrowded training task was different from the single-letter task at pretest, because during the uncrowded training a stimulus jump was involved. The crowded training was similar to the crowded letter task.

In the single-letter task, a central C was presented for 500 ms in one of four orientations: up, down, left, and right. Children had to report the orientation of the C by pressing a corresponding key on the keyboard. Letter size was varied from 0.2 log units below to 0.3 log units above the uncrowded DVA measured during the pretest ophthalmologic assessment (six sizes, 15 trials per size). Feedback was given after each trial by presenting a red or green smiley for 200 ms according to the correctness of the child's response. A Weibull function was fitted to the resulting psychometric response function to find the 62.5% correct performance threshold and the corresponding letter size (in logMAR). This threshold level was chosen because it is halfway between the guess rate (γ, which is 25% for a four-alternative forced-choice task) and the 100% correct rate (see “Data Preprocessing” below).⁴⁹ 

In the crowded letter task, children were instructed to identify a target C surrounded by six Cs of the same size in another orientation. This stimulus array was presented either to the left or to the right of the initial fixation. Letter size was kept fixed at 0.15 log units above the subjects' uncrowded VA measured at baseline with the single-letter task. This enlargement was based on pilots in children with normal vision and lies close to the 0.12 log units enlargement used to measure crowding extent in a previous study.⁵⁰ In each trial, the first stimulus was a center stimulus patch made up of seven Cs all in the same orientation (Fig. 1A, right). The target C had a different orientation than the surrounding Cs, which were placed at 60° intervals around the target. Trial-by-trial feedback about the correctness of the answer was given by a red or green smiley, which was presented for 200 ms. Target-to-flanker distance (in minutes of visual angle, and measured from center to center) was a multiplication of letter height: ×1.2, ×1.5, ×2.0, ×2.5, ×3, and ×4. The 62.5% correct discrimination thresholds were found by fitting Weibull functions to the resulting psychometric response functions (see “Data Preprocessing” below). The resulting spacing thresholds were expressed in minutes of visual angle and then log transformed to obtain crowding extent scores in logMAR: crowding extent = log₁₀(spacing threshold). 

If the 62.5% correct crowding extent could not be found with this fixed-stimulus procedure, an adaptive staircase procedure was used (Quest).⁵¹ The Quest procedure consisted of two parts: a part in which 40 trials were presented to determine the uncrowded VA, and a part in which 40 trials were run to assess the crowding extent for that letter size. This procedure was used in 18/36 children with IN and 5/11 children with normal vision. In these cases, the letters in the crowded letter task had to be enlarged by 0.37 ± 0.04 logMAR and 0.26 ± 0.02 logMAR, respectively. Thus, children with IN needed more enlargement of the target letter to perform the crowded letter task than children with normal vision. 

During the pre- and posttest sessions a chin rest was used to keep children at the appropriate distance from the monitor and an infrared camera, which was used to record their eye movements (for details, see Ref. 51). For children with a DVA > 0.7 logMAR, eye-to-screen distance was 50 cm, and distance between the two groups with Cs was 15° to prevent patches from overlapping. For children with DVA ≤ 0.7 logMAR, eye-to-screen distance was increased to 130 cm and center-to-center distance between the patches was 5° to prevent the second patch from falling outside the screen dimensions and to provide good image quality. 

In addition to these letter-discrimination tasks, all children also performed an eye fixation task, a saccade task, and a reading task to assess their oculomotor behavior and reading performance. A detailed description of these tasks and their outcomes is given elsewhere (see Refs. 52, 53). 

Training

Children were trained on a 15-inch laptop. The crowded training was similar to the crowded letter task except that target size and target-to-flanker spacing were adjusted based on the child's performance (Fig. 1B, right). Children started the first training at the letter size and spacing found at pretest. Spacing was reduced by 0.1 log units if at least 7 of 10 answers were correct; if fewer than 7 out of 10 answers were correct, spacing was increased by 0.1 log units. If letters were closer to each other than 1.1× the letter size, letter size was reduced by 0.1 log units and the target-to-flanker spacing was adjusted to 3× the letter size. 

In the uncrowded training group, children worked with single Landolt Cs instead of groups, but otherwise their task was identical (Fig. 1B, left). That is, the children first had to look at a central C and then identify the orientation of a second, peripheral C that was presented at the left or right side of the center C (for 500 ms). Letter size was reduced by 0.1 log units if at least 7 of 10 answers were correct; otherwise, letter size was increased by 0.1 log units. By presenting single letters, no crowding was provoked. The objectives of implementing a two-step routine in both training paradigms were to the train oculomotor and attention control, and possibly invoke more generalized learning. 

Each training session consisted of seven blocks of 50 trials. In between blocks, children played a reward game (see “Reward Game” below). Children were trained 2× per week during five consecutive weeks, making a grand total of 3500 trials per training. At the end of a training session, the threshold letter size and spacing were determined (crowded training group) or the uncrowded VA was determined (uncrowded training group). Subsequent training sessions always started at the threshold measured during the previous training. During training, viewing distance was 100 cm for children with VA ≤ 0.7 logMAR, 50 cm for children with DVAs between 0.7 and 1.0 logMAR, and 30 cm for children with DVA > 1 logMAR. Distance was monitored by the trainer. 

Reward Game

After each of the seven training blocks, the children played a reward game. Initially, the picture of an animal was masked by a 7 × 7 rectangle matrix covering the whole picture (Fig. 2). The child could click the mouse one to seven times after each training block. The number of mouse clicks depended on the child's performance: more mouse clicks could be earned with better performance. With one mouse click, one part of the picture was revealed. Children were asked to guess what kind of animal they saw after each mouse click. 

Figure 2

 

Example of the game children played between blocks. The number of mouse clicks depended on the child's performance (translated for illustration purposes).

Procedure

The time line in Figure 3 summarizes the experimental procedures for each group of participants. Children with IN were assigned to a training group by letting a computer randomly select a training condition for each child based on age and diagnosis to ensure that these variables were distributed equally across training groups (Table 1). At pretest, ophthalmologic and task-related measures were collected for children with normal vision and children with IN. Children with normal vision performed the computer tasks a second time 7 to 10 days after the first measurement. 

Figure 3

 

Schematic presentation of the study protocol. Children with IN were tested before and after 10 training sessions in which they performed a crowded or uncrowded letter-discrimination task on a laptop computer. Norm data and test–retest values were obtained from children with normal vision. The current paper reports only on training-induced changes in ophthalmologic measures and letter-discrimination performance in the computer tasks. Data from the other tasks are presented in two companion papers.^52,53

Table 1

 

Baseline Measures of Children with Albinism (Alb), Idiopathic Infantile Nystagmus (IIN), and Normal Vision (NV); ANOVAs Performed to Evaluate Differences Between Groups (F Value, P Value, and Post Hoc Pairwise Comparisons Between Groups)

 

Children with IN started their training within 2 weeks after pretest. During training, children were seen twice a week over a period of 5 weeks (10 sessions). Each session consisted of approximately 20 minutes of practice. Trainers visited children at their schools. Within 2 weeks after the last training, children performed the posttest. 

Data Preprocessing

Discrimination thresholds were obtained by fitting Weibull functions to the psychometric response curves obtained in the pre- and posttests using the following equation:  

 

where α represents the discrimination threshold, β the steepness of the slope, γ the guess rate, and λ the lapse rate; λ and γ were set to fixed values of 0.01 and 0.25, respectively. The two free parameters, α and β, were estimated with a maximum likelihood procedure⁴⁹; β was included as free parameter too because of the diversity of the shapes of psychometric response functions across subjects.  

Training thresholds were calculated at the end of each training session. In the crowded training, two variables could be adjusted during training: spacing (the primary variable) and letter size (letter size was adjusted if letters were abutting or if letter spacing was more than 3× the letter size). Therefore, the first step in fitting the psychometric curve for subjects in the crowded training group was to look for the letter size that corresponded best with 62.5% correct performance and use responses only for this letter size to estimate the spacing threshold. If only one letter size was used during a training session, all responses from that training session were used to determine the spacing threshold. For children in the uncrowded training group, all responses in a particular training session could be used since only letter size was adjusted. Thresholds were obtained by fitting Weibull functions to the data as described above. 

Equipment

Stimuli for the pre- and posttests were generated by a 15.6-inch laptop (Dell M4700; Dell Inc., Round Rock, TX, USA) equipped with an OpenGL graphics card (Nvidia Quadro K2000M; Santa Clara, CA, USA) and presented on a 32-inch LCD (Dell UP3214Q; 3840 × 2160 pixels, pixel pitch 0.18 mm²). The training was executed on a laptop equipped with an OpenGL graphics card (Nvidia Quadro K1100M) and a 15.6-inch LCD screen (Dell M3800; 3200 × 1800 pixels, pixel pitch 0.11 mm²). Background luminance of the 32-inch monitor, measured with a luminance meter (Minolta LS-100; Nieuwegein, The Netherlands), was 0.3 cd/m² for a black background and 193.8 cd/m² for a white background. Background luminance of the training laptop was 0.3 cd/m² for a black background and 380.8 cd/m² for a white background. Experimental stimulus software was written in Matlab (version 2014b; MathWorks, Inc., Natick, MA, USA) using the Psychophysics Toolbox (version 3.0.12).⁵⁴ Stimulus timing and button responses were recorded and stored at millisecond precision for offline analysis. 

Statistical Analysis

Statistical analysis was done with SPSS (version 21.0; IBM Corp., Armonk, NY, USA). Baseline performance measures were compared for children with albinism, children with idiopathic IN, and children with normal vision using univariate ANOVAs with Bonferroni correction for multiple comparisons. In addition, we checked equality of training groups with an independent samples t-test. Training effects (and test–retest learning) were evaluated with a repeated measures ANOVA with training group entered as the between-subjects variable and pre- and posttest as the within-subjects variable. Diagnosis did not affect training effects and was therefore not included as a factor. In case PREPOST × training-type interactions occurred, post hoc repeated measures ANOVAs were run to evaluate training effects for the two training groups separately. Unless otherwise specified, interaction effects were absent. Results were considered statistically significant if alpha (type I error) was <0.05. 

Results

Baseline Differences Between Groups

Table 1 lists the baseline differences between children with albinism (all having IN), children with idiopathic IN, and children with normal vision. As expected, VAs and stereopsis were poorer and crowding intensity and extent were larger for children with IN than for children with normal vision. In addition, baseline acuities and crowding extent were larger in children with albinism than in children with idiopathic IN. Table 2 lists the baseline differences between the two training groups. Note that were no significant differences between the uncrowded training group (n = 18; albinism: n = 8; idiopathic IN: n = 10) and crowded training group (n = 18; albinism: n = 10; idiopathic IN: n = 8) on any of the outcome measures. 

Table 2

 

Baseline Visual Performance Measures for Children With IN in the Crowded Training Group and Children in the Uncrowded Training Group

 

Training

Figure 4 displays learning curves for both training groups. Data from individual participants were normalized to allow for averaging across subjects. Figure 4A displays the normalized letter-size threshold, which was calculated by subtracting the pretest uncrowded VA (in logMAR) established with the single-letter task from the letter-size threshold in a particular training session (in logMAR). Note that there was a systematic decrease in letter size in both training groups over the course of the training. In the uncrowded training group (black squares), the letter-size thresholds in the first training session were notably higher than the single-letter discrimination threshold measured at pretest (i.e., normalized size > 0). This can be explained by the stimulus jump that was included in the uncrowded training paradigm but not in the single-letter task, which meant that the children now had to redirect their gaze before they could scrutinize the target. After two training sessions, children in the uncrowded training group typically worked with letter sizes that were below their pretest threshold, indicating that they got quickly accustomed to making a saccade toward the target C at the left or right of the center stimulus. Subsequent progress was much slower, but still statistically significant as indicated by a linear trend analysis on the data from sessions 4 to 10 (correlation r = −0.22, P < 0.02). Letter sizes were systematically larger in the crowded training group (red circles). This is indicative of the degrading effect of nearby distractors on target identification (i.e., crowding). Also note that children in the crowded training group always trained with letters well above their uncrowded VA (i.e., normalized size > 0), which means that the training did not push for improvements in uncrowded VA. 

Figure 4

 

Averaged learning curves. (A) Normalized letter-size threshold in logMAR during the training sessions in the crowded (red circles) and uncrowded training group (black squares). Note the systematic improvement in task performance over the course of the training sessions in both groups. (B) Normalized spacing threshold of the letters in logMAR as a function of training session number. Crowded training group only. Thresholds were estimated from the data collected during training (Methods). Error bars: ±1 SEM.

Figure 4B displays normalized letter spacing as a function of training session. It shows only data from the crowded training group (red circles) since these were the only children trained under crowded conditions. Notice the steep decrease in spacing over the course of the first three sessions (Fig. 4B) with little change in letter size (Fig. 4A), followed by a slower decrease in letter size and spacing over the course of subsequent sessions. Spacing threshold did drop significantly between sessions 3 and 10 (paired t-test: t = 3.08, P < 0.01), but due to the concomitant reductions in letter size, this decrease did not follow a smooth trajectory. The gradually subsiding effect of nearby distractors after the third crowded training session is nevertheless evident from the significant decrease in letter-size threshold over the remaining sessions (linear trend analysis: r = −0.27, P < 0.001). 

Task-Specific Changes After Training

As suggested by the learning curves, training induced robust improvements in task performance. Training effects were the same for children with albinism and idiopathic IN and therefore pooled in the analyses. The following paragraphs quantify these changes. 

Single-Letter Computer Task.

Figure 5A displays the average pre- and posttest uncrowded DVA measured in the single-letter computer task together with the averaged difference between pre- and posttest scores (diff). There were four missing values caused by technical problems (ID, where ID refers to identification number, 1, 2, and 7) or lack of cooperation (ID 5). Note that uncrowded VAs were equally reduced in both training groups after training (PREPOST: F(1,30) = 35.01, P < 0.001, partial η² = 0.54). Mean improvement was 0.085 ± 0.014 logMAR. 

Figure 5

 

Changes in computer-task measures. (A) Uncrowded VA as measured with the single-letter task, and (B) crowding extent as measured with the crowded letter task. Left and center show data from children with IN in the crowded and uncrowded training groups before and after their training. Right shows data from the first and second measurement in children with NV. Paired differences (diff) indicate the mean change between test sessions. Note improved task performance after training. Error bars: ±1 SEM. ***P < 0.001.

Crowded Letter Computer Task: Crowding Extent.

Figure 5B shows means and paired difference of crowding extent at pre- and posttest. After training, crowding extent was 0.25 ± 0.04 logMAR smaller than at baseline (PREPOST: F(1,34) = 47.34, P < 0.001, partial η² = 0.60). Remarkably, uncrowded and crowded training appeared to have a very similar effect on crowding extent; no group effects or interactions were found (P > 0.1). 

Test–Retest Learning.

Test–retest learning was examined in a group of normal-vision children (Figs. 5A, 5B, right). In these children, uncrowded VA did not differ between the first and second measurement (F(1,10) = 2.84, P = 0.123, partial η² = 0.22). Crowding extent was also not significantly different between the first (0.39 ± 0.03 logMAR) and second measurement (0.32 ± 0.04 logMAR) (F(1,9) = 4.01, P = 0.076, partial η² = 0.31). 

Before–After Differences Ophthalmologic Exam

Subsequent analysis of the ophthalmologic exam scores showed that our training protocols also induced robust improvements in visual acuity and stereopsis as well as specific reductions in crowding. 

Binocular Distance Visual Acuity.

On average, training significantly improved the uncrowded DVA in both training groups by 0.10 ± 0.02 logMAR (PREPOST: F(1,34) = 25.52, P < 0.001, partial η² = 0.43; Fig. 6A). Crowded DVA also showed an equal improvement for both training groups of 0.11 ± 0.02 logMAR (PREPOST: F(1,34) = 29.76, P < 0.001, partial η² = 0.47; Fig. 6B). 

Figure 6

 

Changes in distance visual acuity. (A) Uncrowded distance visual acuity (DVA) and (B) crowded DVA at pre- and posttest together with the corresponding before–after differences in the crowded and uncrowded training group. Note DVA improvements in both groups. Error bars: ±1 SEM. ***P < 0.001.

Monocular Distance Visual Acuity.

Improvements in monocular distance visual acuities paralleled the ones seen for binocular DVA in both training groups (see Supplementary Table S2). 

Binocular Near Visual Acuity.

Uncrowded NVA improved equally in both training groups by 0.07 ± 0.02 logMAR (PREPOST: F(1,34) = 18.82 P < 0.001, partial η² = 0.36; Fig. 7A). Improvements for crowded NVA did differ between groups (PREPOST × training group: F(1,34) = 6.41, P = 0.016, partial η² = 0.16; Fig. 7B). Crowded NVA improved by 0.15 ± 0.02 logMAR for the crowded training group (PREPOST: F(1,17) = 36.05, P < 0.001, partial η² = 0.68), but did not change in the uncrowded training group (0.05 ± 0.03 logMAR; PREPOST: F(1,17) = 2.91, P = 0.106, partial η² = 0.15). 

Figure 7

 

Changes in near visual acuity. (A) Uncrowded near visual acuity (NVA), and (B) crowded NVA at pre- and posttest together with the corresponding before–after differences in the crowded and uncrowded training group. Note absence of crowded NVA improvement in the uncrowded training group. Error bars: ±1 SEM. ***P < 0.001.

Binocular Distance Crowding Intensity.

The distance crowding intensity was not significantly altered in either training group (PREPOST: F(1,34) = 0.51, P = 0.480, partial η² = 0.02). Distance crowding intensity was 0.198 ± 0.014 logMAR at pretest and 0.183 ± 0.019 logMAR at posttest (Fig. 8A). 

Figure 8

 

Changes in crowding intensity. (A) Distance crowding intensity (DCI), and (B) near crowding intensity (NCI) at pre- and posttest together with the corresponding before–after differences in the crowded and uncrowded training group. Note reduced NCI after training under crowded conditions. Error bars: ±1 SEM. **P < 0.01.

Binocular Near Crowding Intensity.

Changes in near crowding intensity differed between training groups (PREPOST × training group: F(1,34) = 5.43, P = 0.026, partial η² = 0.14; Fig. 8B). Near crowding intensity in the uncrowded training group was unaltered after training (PREPOST: F(1,17) = 0.30, P = 0.592, partial η² = 0.02), but there was a significant reduction of the near crowding intensity in the crowded training group (PREPOST: F(1,17) = 10.00, P = 0.006, partial η² = 0.37). It was reduced by 0.08 ± 0.03 log units. 

Stereopsis.

Stereopsis showed an equal improvement in both training groups after training (PREPOST: F(1,30) = 11.26, P = 0.002, partial η² = 0.27; Fig. 9). It changed from 993 ± 341 arcsec to 493 ± 158 arcsec in the uncrowded training group and from 1082 ± 363″ to 218 ± 168″ in the crowded training group. Mean stereopsis improvement was 500 ± 365 arcsec in the uncrowded training group and 863 ± 340 arcsec in the crowded training group. 

Figure 9

 

Changes in stereopsis. Stereopsis (log₁₀ transformed) before and after training together with the corresponding before–after differences in the crowded and uncrowded training group. Note improved stereopsis in both training groups. Error bars: ±1 SEM. ***P < 0.001.

Discussion

Our results show that a modest amount of perceptual training not only causes robust, task-specific performance improvements, but also induces improvements in distance and near visual acuity and even in stereopsis. Training under crowded conditions resulted in significantly larger improvements with regard to crowded near visual acuity and the near crowding intensity than training under uncrowded conditions. 

Task-Specific Training Effects

Training improved task performance in both groups (Fig. 5). The improvement on the single-letter computer task (0.085 ± 0.014 logMAR, partial η² = 0.54) is nicely in line with and extends recent studies reporting VA improvements of 0.085 to 0.15 logMAR after a short period of perceptual learning in adults with amblyopia.^38,39 This task-specific improvement was expected for the uncrowded training group, because these children trained with near-threshold letter sizes, but not so much for the crowded training group. Although letter sizes were gradually reduced in the crowded training too, they were always well above the child's initial uncrowded VA (Fig. 4). Conversely, we found that the uncrowded training group showed a large improvement on the crowded letter task even though these children did not train under crowded conditions. In fact, the reduction in crowding extent was about the same in the uncrowded (0.20 ± 0.05 logMAR) and crowded training groups (0.30 ± 0.05 logMAR). Our findings thus indicate near-complete transfer of learning between the two computer tasks. 

Several observations indicate that the improved task performance did not reflect simple test–retest learning effects. First, it is clear from the learning curves (Fig. 4) that performance improved gradually and that only the letter-size reductions achieved toward the end of the training closely matched the improvements in uncrowded VA (Fig. 5). Second, children with normal vision performed the computer tasks equally well on both test occasions. It is possible that the absence of a significant increase in their uncrowded VA might reflect a ceiling effect because their initial VA was already very good (−0.31 ± 0.08 logMAR), but a ceiling effect is not expected for crowding extent; in children with normal vision, crowding decreases with age until at least the age of 11 years.⁵⁰ Third, the ophthalmologic acuity tests, for which test–retest reliability has been established,^55,56 also demonstrated clear improvements in both crowded and uncrowded visual acuity, and these improvements were strongly correlated with the improvements in task performance. 

Training Effects on Ophthalmologic Measures

Task-related learning effects generalized to improved mono- and binocular uncrowded and crowded VA as measured with standard VA charts (Figs. 7, 8). In fact, the DVAs were very consistent across computer- and experimenter-guided assessments (correlation: r = 0.96, P < 0.001). Since computer acuity assessments are not influenced by the experimenter, it is unlikely that the reported VA improvements resulted from potential experimenter bias. It also unlikely that the effects of training on the ophthalmologic measures were due to repeated testing; all participants with IN were already highly familiar with the visual acuity charts before entering this study because of regular assessment by their practitioner. We cannot exclude the possibility that acuity improvements are partly due to perceived benefit of the training, but this explanation seems at least incomplete since the acuity improvements reported in our study, which range from 0.085 to 0.15 logMAR, are significantly larger than the 0.04 ± 0.03 logMAR acuity improvements that were previously found in adults with IN after a placebo treatment.⁵⁷ 

Other studies with successful generalization of learning after perceptual learning have also reported improved VA after repeated training with near-threshold letter sizes.⁵⁸ As far as we know, the present study is the first to demonstrate improved DVA and NVA after perceptual learning in children with IN. Training even improved stereopsis by almost 1.5 octaves. This improvement is quite remarkable because in normal visual development this change in stereopsis corresponds to an improvement seen between 24 and 36 months,⁵⁹ indicating that our training induced a 1-year catching-up in stereopsis development. To our knowledge, such beneficial generalization of perceptual learning has not been reported before. 

The key difference between our two training paradigms was their ability to improve crowded near vision. While the uncrowded NVA improved considerably after both types of training, crowded NVA improved significantly only after training under crowded conditions (Fig. 8). Near crowding intensity was also specifically influenced by the crowded training (Fig. 9). Improved NVA after a short period of perceptual learning with a paper-and-pencil paradigm was reported earlier by our group.⁶⁰ 

Training Reduces the Near, but Not Distance, Crowding Intensity

While the near crowding intensity was reduced in the crowded training group (0.08 ± 0.03 logMAR, partial η² = 0.37), the distance crowding intensity did not change. Participants did show large improvements in crowded DVA after training (0.11 ± 0.02 logMAR), but even in the crowded training group the uncrowded DVA improved by an almost similar amount (0.10 ± 0.02 logMAR). The same effect was also reported in a recent study using perceptual learning to improve foveal crowded vision in presbyopes where not just crowded vision, but also untrained uncrowded vision improved.^7,58 The lack of a reduction in distance crowding intensity seems inconsistent with an earlier study showing reduced crowding intensities after 8 to 14 sessions of crowded perceptual learning.⁶ However, different target–flanker spacings were used in that study. Crowding intensity decreased after training only at the largest spacing (1.4× optotype size), not for the smallest spacings (1.1× and 1.2× optotype size), which were closer to the spacing used in our study. One subject in the Hussain study continued training with small spacings for 11 additional sessions, resulting in a further reduction of crowding.⁶ This suggests that it takes longer to reduce crowding intensity if subjects are trained with small optotype spacings. We therefore believe that our 5-hour training period might have been too short to reduce the distance crowding intensity in children with IN. Perhaps the distance crowding intensity is more difficult to reduce than the near crowding intensity, because baseline near crowding intensities were higher than those at distance and there is more room for improvement when impairments are larger.^61,62 

Possible Mechanisms of Improvement

Sensory aspects (low-level improvements associated with great specificity⁶³), cognitive aspects (better recognition due to rule learning/high-level learning⁶⁴), and oculomotor aspects may all have contributed to the observed improvements. For example, recent brain imaging studies show that perceptual learning can modify responses in early visual cortices,^65,66 and induce changes in gray-matter volume in function-related cortical areas.⁶⁷ In the context of our present study, changes in visual areas V1 to V4 are likely, since these areas show suppressed activity when crowding occurs^68,69 and reduced population receptive field sizes after perceptual learning.^70,71 Albinism is often accompanied by foveal hypoplasia and hypoactivity in the posterior pole of the visual cortex representing the fovea.⁷² Perhaps perceptual learning helps in restoring the activity levels in the posterior pole of V1. 

In adults with normal vision, it was reported that perceptual learning improves VA and reduces visual processing time.²⁴ Reduced processing time could be an explanation for the improvements in visual acuity in children with IN because they show shorter foveation periods and limited processing time.^19,73 In general, if attention is directed at the target stimulus, learning is more likely to be high level. By contrast, if learning is passive and no attention is directed at the stimulus, changes on a low level seem to be greater.³² 

Fast learning occurred during the first three training sessions and was followed by slow learning. A possible explanation for the rapid improvements in task performance during the first training sessions is that subjects were getting familiar with the task structure. This type of learning is called procedural learning.^34,74,75 If the net training effects were to be due purely to procedural learning, the effects should be the same for the crowded and uncrowded training groups, because the two-step experimental task procedure was similar for our training groups. There were, however, differences in training effects; the crowded training group showed larger improvements on the (crowded) near vision tasks than the uncrowded training group. Furthermore, we observed improvements on tasks that were very different from the trained task, namely, visual acuity improvements, stereopsis improvements, and, as we report elsewhere,^52,53 improvements in reading performance. These improvements were all well above test–retest changes in normal controls and exceeded previously reported placebo effects.^52,53 We therefore think that the considerable transfer across tasks that we saw is evidence of genuine improvements in visual perception rather than the result of an improved familiarity with the experimental task demands. 

It is unlikely that the visual performance improvements reported in the present study are due to improved oculomotor control either, since we did not find training-induced changes in two-dimensional oculomotor behavior.⁵² Previous studies have reported that training paradigms involving the execution of saccades of a specific amplitude and direction can increase the occurrence of express saccades,⁷⁶ which would allow longer target inspection times.⁵⁷ This was not the case; training did not result in changes in saccade latency and/or gain, nor were there changes in fixation stability.⁵² One aspect of oculomotor behavior that might account for some of the transfer of learning effects is improved binocular coordination. The oculomotor component of the training also enforced proper control of vergence eye movements to realign the fovea of each eye with the peripheral target. Thus, it is possible that better binocular coordination might account for the stereopsis improvement.⁷⁷ 

Conclusions

Children with IN show higher foveal crowding intensities than children with normal vision. Ten sessions of perceptual learning on a crowded letter identification task resulted in reduced crowding intensities at near, improved near and distance crowding acuity, and improved stereopsis. The larger progress in the crowded training group compared to the uncrowded training group indicates that training children to identify a target among distractors can significantly improve visual functions. Longer training might induce larger improvements. 

Acknowledgments

The authors thank the parents and children for their participation. In addition, we thank Anne-Lotte Nijhof, Kim Ligtenbarg, Carmen Lageweg, and Debbie Günes-Huurneman for training the children. Finally, we thank Marlijn Anneveldt for running a preliminary analysis of this study. 

Supported by the ODAS Foundation and Landelijke Stichting voor Blinden en Slechtzienden (LSBS) which contributed through UitZicht, and Bartiméus Sonneheerdt. The funding organizations had no role in the design or conduct of this research. 

Disclosure: B. Huurneman, None; F.N. Boonstra, None; J. Goossens, None