Monocular Contrast Sensitivity Visual Perceptual Learning Rebalances Adult Amblyopes’ Two Eyes

Abstract

Purpose: This study investigates the effects of monocular contrast sensitivity visual perceptual learning on binocular visual functions of adults with binocular imbalances.

Methods: Sixteen adults with anisometropic amblyopia (mean age, 24.63 ± 3.56 years), 20 adults with myopic anisometropia (mean age, 24.20 ± 1.94 years), and 16 visually normal adults (mean age, 24.88 ± 1.89 years) participated in this study. Each group was evenly divided into training (anisometropic amblyopia, myopic anisometropia, normal training group) and untrained control groups (anisometropic amblyopia controls, myopic anisometropia controls, and normal controls). Training groups underwent 10 days of monocular contrast sensitivity perceptual learning (two-alternative forced-choice contrast detection task at 6 cycles per degree) using the amblyopic or nondominant eye, whereas the control groups received no intervention during the same period. Monocular visual acuity, monocular and binocular contrast sensitivity, and balance point were measured before and after the intervention.

Results: Monocular contrast sensitivity perceptual learning significantly improved both contrast sensitivity and balance point at the trained spatial frequency (6 cycles per degree) in the trained eyes of the anisometropic amblyopia training group and myopic anisometropia training group, with improvements generalizing to nearby untrained spatial frequencies. However, no significant improvements were observed in binocular summation ratios for either group. The normal training group showed modest improvements limited to the trained eye at both trained and neighboring spatial frequencies, without significant binocular or untrained eye benefits.

Conclusions: Monocular contrast sensitivity visual perceptual learning effectively enhances monocular visual performance and positively affects binocular functions across trained and nearby untrained spatial frequencies, indicating its potential clinical usefulness in improving binocular vision among adults with binocular imbalances.

Amblyopia, also known as lazy eye, is a developmental visual impairment caused by abnormal visual experiences during childhood, affecting approximately 2% to 5% of the population, with projections suggesting it could impact 220 million people globally by 2040.^1–3 Unlike other visual impairments, amblyopia is characterized by functional deficits in the absence of detectable structural abnormalities in the visual pathway. It typically results from unequal refractive errors (anisometropia), ocular misalignment (strabismus), or visual deprivation, each of which interferes with normal visual development.⁴ The amblyopic eye (AE) suffers from poor vision, including decreased visual acuity and contrast sensitivity, which cannot be corrected optically.⁵ Patching of the fellow eye (FE) is considered the gold standard for amblyopia treatment clinically.^6,7 However, patching therapy is less effective for older children beyond the critical period of visual development (typically >8 years old). For adults >17 years of age, the effectiveness is minimal, making the treatment of adult amblyopia a significant challenge.⁴ 

Contrast deficits are considered a key factor contributing to the poor spatial vision of the AE.⁸ The amblyopic visual system may be noisier and less efficient in processing contrast.⁹ Based on this understanding, recent studies have developed specialized visual perceptual learning tasks—extensive, repetitive, and simple visual exercises—to improve contrast processing in adults with amblyopia. For example, Polat et al.¹⁰ demonstrated that perceptual learning using a flankered contrast detection task significantly improved contrast sensitivity and visual acuity in adult amblyopes. Similarly, Zhou et al.¹¹ demonstrated that hundreds of trials of contrast detection training at cutoff spatial frequencies (a relatively high frequency at which the contrast threshold was 0.5 from pre-training contrast sensitivity measurements) improved the contrast sensitivity in adolescents and adults with amblyopia by approximately 4.9 dB (or 76.5%), with corresponding improvements in visual acuity of 4.5 dB (68.4%). These training effects were maintained for 6 months to 2 years after training.¹¹ Compared with normal adults, amblyopes exhibited broader transfer effects to untrained spatial frequencies after training at a given cutoff spatial frequency.¹² This finding suggests that the visual system of adults with amblyopia retains a degree of plasticity that can be harnessed through contrast sensitivity perceptual learning. Further studies indicate that this training enhances contrast sensitivity in neurons of the primary visual cortex (V1) in cats¹³ and increases eye- and hemifield-specific responses to low-contrast patterns in the magnocellular layers of the lateral geniculate nucleus (LGN) using functional magnetic resonance imaging in adults,¹⁴ with the induced plasticity resulting from increased contrast gain of the trained neurons. These findings suggest that targeting contrast perception in the AE through perceptual learning could be a potential therapeutic approach for adults amblyopia. 

Despite the significant monocular deficits in the AE, amblyopia also affects binocular visual processes.¹⁵ Many studies have documented anomalies in binocular vision in amblyopia, including abnormal binocular combinations,^16,17 abnormal interocular interactions,¹⁸ and stereoscopic anomalies.¹⁹ Recent studies suggest that patching therapy, although effective in improving visual acuity in the AE, does not restore the binocular balance between the eyes under binocular viewing conditions.²⁰ Amblyopes may continue to experience binocular imbalance even after their AE has achieved normal visual acuity.² Given that we perceive the world through both eyes, these binocular vision abnormalities can pose significant challenges in daily life for individuals with amblyopia, affecting fine motor control, postural stability, and performance in activities such as sports and driving.^21,22 These studies underscore the importance of addressing binocular impairment in the treatment of amblyopia. 

Our understanding of binocular impairment in amblyopia is gradually evolving. Studies suggest that these visual function impairments may be related to deficits in contrast processing. Specifically, decreased contrast sensitivity in the AE and the imbalance in contrast sensitivity between the two eyes contribute to binocular dysfunction. Adjusting contrast to compensate for these differences can help to restore binocular balance. This effect is consistently observed in measurements like binocular phase combination²³ and binocular orientation combination,²⁴ as well as binocular rivalry.¹⁶ Furthermore, adjusting the contrast of the monocular inputs can be potentially applied for improving stereopsis in individuals with binocular imbalance.²⁵ Amblyopia also involves temporal processing impairments, with interocular delays being mediated by interocular contrast difference.²⁶ Based on these findings, several models, such as the two-stage model,²⁷ the Ding-Sperling model,²³ and the multipathway contrast-gain control model,²⁸ have been proposed to elucidate binocular deficits in amblyopia and their connection to contrast processing. 

Drawing from these studies, one intriguing hypothesis is that contrast sensitivity perceptual learning in the AE might improve binocular visual functions in amblyopia by enhancing contrast processing. This hypothesis was tested in two recent studies by Chen et al.²⁹ and Jia et al.³⁰ Both studies reported enhancements in stereopsis and other binocular functions after monocular contrast sensitivity visual perceptual learning. However, their findings also revealed some discrepancies. To illustrate, whereas Chen et al. found that training helped to rebalance amblyopes’ two eyes in binocular phase combination tasks, Jia et al. did not observe similar results. It is worth noting that the trained frequency for the AE (e.g., 14.9 cycles per degree [cpd], corresponding with the cutoff spatial frequency of the contrast sensitivity function before training) is relatively high and the frequency used to measure binocular function (e.g., 0.293 cpd) in these studies are relatively low. These spatial frequencies were different, with a disparity of up to 6 octaves. Huang et al.¹² suggested that, although amblyopic subjects exhibit broader transfer effects from trained to untrained frequencies compared with normal observers, the average full bandwidth of the transfer is approximately 4 octaves (compared with 1.40 octaves for normal observers), indicating that the transferability of perceptual learning may be limited by bandwidth constraints. Thus, the response to whether monocular contrast sensitivity perceptual learning can affect binocular visual functions is not straightforward and clear. A more direct approach would be to train participants on contrast detection tasks at a specific frequency and then assess binocular functions at both the trained and untrained frequencies. 

In this study, we aimed to provide a clear test for this hypothesis. Specifically, we conducted monocular contrast detection training at a specific spatial frequency (i.e., 6 cpd) in participants with anisometropic amblyopia and myopic anisometropia of varying degrees of binocular imbalance. The inclusion of a myopic anisometropia group as a control was based on our prior findings that these individuals exhibit binocular imbalance similar to amblyopes,^31,32 indicating the potential for improvement in binocular visual function. Additionally, lenses used to correct anisometropic amblyopia and myopic anisometropia can distort the perceived shape and size of images between the two eyes, which may influence the generalizability of training effects. In this sense, such distortions could provide a broader range of visual stimuli compared with controls without glasses or anisometropia. Furthermore, we also included a group of normal participants and assigned all participants (anisometropic amblyopia, myopic anisometropia, and normal vision) to either a training or untrained control group to enable a clearer comparison of training effects across different participant types. We measured binocular function indicators related to contrast processing, focusing on both suprathreshold (balance point) and threshold-level (binocular summation ratio) indicators at the trained frequency and several untrained frequencies (1.5, 3.0, and 8.0 cpd) before and after training. We then evaluated changes in contrast sensitivity at the trained frequency in the trained eye (AE/nondominant eye [nonDE]) and assessed the impact of monocular contrast sensitivity perceptual learning on both balance point and binocular summation ratio at the trained frequency, as well as its effects on untrained spatial frequencies. 

Methods

Participants

Sixteen adults with anisometropic amblyopia (mean age, 24.63 ± 3.56 years), 20 with myopic anisometropia (mean age, 24.20 ± 1.94 years), and 16 normal adults (mean age, 24.88 ± 1.89 years) participated in the study. Each group was evenly divided, with one-half assigned to the training groups (anisometropic amblyopia training group [ATs], myopic anisometropia training group [MTs], normal training group [NTs]) and the other one-half to the untrained control groups (anisometropic amblyopia untrained control group [ACs], myopic anisometropia untrained control group [MCs], normal untrained control group [NCs]). The mean interocular spherical equivalent error difference was 4.14 ± 2.80 D in the ATs and 4.81 ± 2.11 D in the ACs. For the myopic anisometropia training group, the spherical equivalent error difference was 2.25 ± 1.28 D, and 1.91 ± 0.48 D in the MCs. In the normal training group and NCs, the spherical equivalent error differences were 0.34 ± 0.30 D and 0.27 ± 0.39 D. No statistically significant differences in spherical equivalent error were observed between the training groups and their control groups as determined by two-way repeated measures ANOVA (between-subject factor, group [training vs. control]; within-subject factor, eye [AE/nonDE vs. FE/DE]): ATs vs. ACs, F(1,7) = 1.876, P = 0.213, partial η² = 0.211; MTs vs. MCs, F(1,9) = 0.557, P = 0.474; partial η² = 0.058; NTs vs. NCs, F(1,7) = 0.367, P = 0.564, partial η² = 0.050. Detailed baseline characteristics are provided in Supplementary Table S1. During the initial visit, a comprehensive ocular examination was conducted, including assessments of visual acuity, stereoacuity, autorefraction, subjective refraction, slit lamp examination of the anterior segment, and direct ophthalmoscopy of the posterior segment to rule out any organic causes of vision loss. Refractive corrections were determined using cycloplegic refraction. The American Academy of Ophthalmology's Preferred Practice Pattern guidelines^4,33 were used to determine the diagnosis and categorization of subjects based on their initial hospital visit records. Specifically, anisometropic amblyopia was defined as an interocular spherical equivalent refractive error difference of ≥1.00 D between eyes and/or an astigmatic difference of ≥1.50 D, with a minimum interocular best-corrected visual acuity disparity of 0.20 logMAR. Anisometropic patients were defined as having a spherical equivalent refractive error difference of at ≥1.00 D, with a best-corrected visual acuity of >0.0 logMAR in both eyes. All subjects signed an informed consent form before the study. The study adhered to the principles of the Declaration of Helsinki and was approved by the Ethics Committee of Wenzhou Medical University, Eye Hospital. 

Design

All observers underwent a pre-training measurement, a training phase, and a post-training measurement (Fig. 1A). The dominant and nonDEs were determined before the pre-training measurement using a binocular orientation combination task.^{17,32,34–36} During the pre-training and post-training assessments, monocular and binocular contrast sensitivity, balance point, visual acuity, and stereoacuity were measured. These tasks were administered in a random order among subjects. In the training phase, the FE of AE subjects or the DE of anisometropic subjects was occluded with opaque material, and training involved a two-alternative forced-choice contrast detection task with a spatial frequency of 6 cpd in the AE or nonDE. Each participant underwent 10 days of training. 

Figure 1.

 

Experimental procedure and task illustrations. (A) Subjects underwent a pre-training measurement, a training phase (contrast sensitivity visual perceptual learning) and a post-training measurement. Visual acuity, monocular and binocular contrast sensitivity (CS) and balance point (measured through binocular orientation combination tasks) were evaluated both before and after the training. (B) During the perceptual learning, the AE or nonDE of subjects underwent a two-alternative forced-choice contrast detection task at a spatial frequency of 6 cpd. (C) Two sinusoidal gratings were presented binocularly, tilted symmetrically by ±7.1° from the horizontal axis. When two eyes are balanced, the perceived orientation of the fused grating will be at 0°. The grating shown to the AE or nonDE was set at a fixed 50% contrast, while that to the preferred FE or DE varied between 0% and 100% contrast, divided into seven contrast levels ranging from 0 to 2. (D) Balance point was measured across spatial frequencies of 1.5, 3.0, 6.0, and 8.0 cpd. (E) The perceived binocular direction of the fused grating was plotted as a function of binocular contrast (FE/AE or DE/nonDE) and fitted with a cumulative Gaussian distribution. The orange points mark the balance point where both eyes contribute equally.

Visual Function Measurement

Visual Acuity and Stereoacuity

Best-corrected visual acuity was measured at a distance of 4 m using the tumbling Early Treatment Diabetic Retinopathy Study chart in logMAR units before and after training. Stereoacuity was assessed before and after training using Randot Stereotests (Stereo Optical Company, Chicago, IL, USA). 

Contrast Sensitivity Function

Sinusoidal grating stimuli were generated on a computer equipped with Matlab (The MathWorks Corp., Natick, MA, USA) and Psychtoolbox^37,38 3.0.9, and presented on an IBM 21-inch CRT monitor (IBM Corp., Armonk, NY, USA) with a resolution of 640 × 480 pixels at 60 Hz and a background luminance of 43.7 cd/m². A special circuit allowed the display system to generate a 14-bit grayscale resolution and perform gamma correction.³⁹ Stimuli were observed monocularly at a distance of 2.88 m under dim lighting on a chin rest. Contrast sensitivity was defined as the reciprocal of the contrast threshold at 79.3% accuracy, measured using a three-down one-up staircase two-alternative forced-choice contrast method. Each stimulus was presented in either a horizontal or vertical orientation, and subjects were required to judge its orientation. After three consecutive correct responses, the target contrast was decreased by 10% (C_n+1 = 0.90C_n), and increased by 10% (C_n+1 = 1.10C_n) after each incorrect response. Contrast sensitivity function was measured at six spatial frequencies before and after training: 0.75, 1.50, 3.00, 6.00, 8.00, and 16.00 cpd. The order of eye testing and spatial frequencies was randomized, and each measurement included 6 blocks of 88 trials, for a total of 528 trials. Practice trials were conducted before the official test to ensure subjects understood the task. 

Binocular Summation

The binocular summation ratio is defined as the ratio of contrast sensitivity values measured under binocular conditions to those measured under the better eye (FE or nonDE) conditions.  

\begin{eqnarray*}{\rm{Binocular\ summation\ ratio}} = \frac{{{\rm{Contrast\ sensitivit}}{{{\rm{y}}}_{{\rm{binocular}}}}}}{{{\rm{Contrast\ sensitivit}}{{{\rm{y}}}_{{\rm{better\ eye}}}}}}\end{eqnarray*}

 

Binocular Orientation Combination Task

Stimuli were generated using Matlab (The MathWorks Corp., Natick, MA, USA) and Psychtoolbox 3.0.1 on an iMAC (Apple Inc., Cupertino, CA, USA) and displayed dichoptically via gamma-corrected head-mounted goggles (GOOVIS, NED Optics, Shenzhen, China).^{17,32,34–36} The goggles had a refresh rate of 60 Hz, a resolution of 2560 × 1600 pixels, and a maximum brightness of 150 cd/m². The primary outcome, termed the balance point, was the interocular contrast ratio (DE/nonDE) at which both eyes contributed equally during the binocular combination (Fig. 1C). A balance point value closer to 1 means less binocular imbalance, whereas greater deviations from 1 signify increased binocular imbalance. Specifically, a balance point value of <1 suggests a stronger DE, and a value of >1 indicates a stronger nonDE. In this study, we measured the balance point at four spatial frequencies (1.5, 3.0, 6.0, and 8.0 c/d) using the method of constant stimuli. Specifically, two horizontally tilted sinusoidal gratings with different orientations were presented in each trial. Each trial tested one of two configurations in a randomized order: in the first configuration, the DE's or FE's grating was tilted counterclockwise (+7.1°) from the horizontal, while the nondominant or AE saw a clockwise tilt (−7.1°); in the second configuration, the orientations were reversed. The total orientation difference between the two eyes was 14.2°. The grating contrast in the nonDE (or AE) was fixed at 50%, and the FE's (or DE's) contrast varied from 0% to 100%. Seven interocular contrast ratios ranging from 0 to 2 were tested, with each condition (a specific orientation and interocular contrast ratio) repeated 20 times, totaling 280 trials per block (2 orientations × 7 contrast ratios × 20 repetitions). Contrast ratios and configurations were randomized across trials, and balance point measurements at different spatial frequencies were randomized across observers. Practice trials were provided to ensure participants understood the task. 

Perceptual Learning

During the perceptual learning, the AE or nonDE of each subject underwent a two-alternative forced-choice contrast detection task at a spatial frequency of 6 cpd. Training used a single extended staircase per day to track each observer's contrast threshold, the starting contrast each day was set to the final value of the previous session, and most trials remained near threshold levels to facilitate perceptual learning. Each participant in the training group received 10 days of training. 

Statistical Analysis

All analyses were conducted using IBM-SPSS 26.0 (IBM Inc.) and data visualization was performed with MATLAB. Normality was assessed using the Shapiro–Wilk test (P > 0.05 indicated a normal distribution). Based on the distribution of data, all pre-training and post-training measurements were compared using paired t tests or paired nonparametric tests. To analyze the effect of monocular perceptual learning on binocular functions across a range of spatial frequencies, a two-way repeated measures ANOVA was conducted with time (before measurement, after measurement) and spatial frequency as within-subject factors. For the linear fitting and normalization analysis of contrast sensitivity, balance point, and binocular summation ratio, data were normalized in decibels, calculated as 20 × log₁₀(X), where X denotes the measurements of contrast sensitivity, balance point, or binocular summation ratio. A P value of <0.05 was considered statistically significant. 

Results

Perceptual Learning Enhances Visual Functions of the Trained Eye at the Trained Frequency

We trained the AE of ATs and the nonDE of MTs and NTs at a spatial frequency of 6 cpd. As shown in Figure 2, this training led to improvements in both contrast sensitivity and the balance point at the trained spatial frequency. However, no significant improvement was observed in the binocular summation ratio. 

Figure 2.

 

Visual function outcomes at the trained spatial frequency. (A) Learning curves for the ATs, MTs, and NTs, illustrating contrast sensitivity as a function of training sessions. Contrast sensitivity measurements before and after training (first and last data points) along with the average across all 10 sessions are displayed. Red circles with error bars indicate the mean ± SE for ATs, blue circles for MTs, and green circles for NTs. Solid lines represent linear fits of the group means. (B) Changes from baseline in contrast sensitivity at the trained spatial frequency (6 cpd) for the trained eye (AE or nonDE) across the three training groups (ATs, MTs, and NTs) and their respective control groups (ACs, myopic anisometropia control group [MCs], and normal control group [NCs]). (C) Changes from baseline in balance point (BP) at 6 cpd for all six groups, measured using the binocular orientation combination task. (D) Binocular summation ratio at 6 cpd for all six groups. Each circle in (B), (C), and (D) represents an individual participant, and error bars denote standard error of the mean (SE). Statistical significance is indicated as follows: *** P < 0.001; **0.001 < P < 0.01; *0.01 < P < 0.05; n.s., not significant.

Monocular Contrast Sensitivity

Perceptual learning significantly enhanced the contrast sensitivity of the trained eye in all three trained groups at the trained spatial frequency (6 cpd). In ATs, the contrast sensitivity of the AE increased from 15.17 to 25.53, resulting in an improvement of 6.27 ± 3.63 dB, corresponding with a 168.26% enhancement (Z = −2.521; P = 0.012). In MTs, the contrast sensitivity of the nonDE improved from 56.20 to 82.08, with a gain of 3.69 ± 2.63 dB, marking a 146.05% improvement (t = −4.298; P = 0.002). Similarly, NTs showed a significant increase in the nonDE (from 33.40 to 52.26), yielding an improvement of 4.28 ± 3.16 dB (156.44%; t = −4.224; P = 0.004). The learning curves of all three trained groups (ATs, MTs; and NTs) were fitted with a logarithmic-logarithmic linear function, as illustrated in Figure 2A. The fitted slopes were 0.26 for ATs, 0.15 for MTs, and 0.17 for NTs, indicating the respective rates of improvement during the training period. In contrast, control groups (ACs, MCs, and NCs) who underwent only pretest and post-tests without any training showed no significant changes in contrast sensitivity of the trained eye at 6 cpd. Specifically, ACs exhibited a change from 18.42 to 17.48 (t = 0.427; P = 0.682), MCs from 39.62 to 47.41 (t = −1.169; P = 0.273), and NCs from 42.44 to 43.27 (t = −0.210; P = 0.840). 

To further compare the changes across all six groups, the change from baseline in the trained eye's contrast sensitivity at 6 cpd was calculated and is presented in Figure 2B. The results demonstrate that the improvements observed in the trained groups were significantly greater than those in the control groups (ATs vs. ACs, t = −4.438, P < 0.001; MTs vs. MCs, Z = −1.965, P = 0.049; NTs vs. NCs, t = −2.237, P = 0.042). 

Binocular Functions

Monocular contrast sensitivity perceptual learning also affected binocular balance at the trained spatial frequency (6 cpd). In ATs, the balance point significantly increased from 0.056 ± 0.075 to 0.085 ± 0.13 (Z = −2.240; P = 0.025). Similarly, MTs showed a significant improvement in balance point, increasing from 0.64 ± 0.13 to 0.73 ± 0.15 (t = −3.564; P = 0.006). In contrast, NTs demonstrated a nonsignificant increase, with the balance point rising from 0.82 ± 0.10 to 0.89 ± 0.09 (t = −1.532; P = 0.169). These findings suggest that monocular contrast sensitivity perceptual learning significantly enhanced binocular balance in ATs and MTs, but had limited effects in NTs. To further compare the improvements between trained and control groups, the change from baseline in balance point was calculated and is presented in Figure 2C. The balance point increase in ATs was significantly greater than in ACs (t = −2.846; P = 0.013). MTs also showed a marginally significant advantage over MCs (t = −1.799; P = 0.089), whereas no significant difference was observed between NTs and NCs (t = −1.289; P = 0.218). 

The binocular summation ratio in ATs increased from 1.23 ± 0.32 to 1.50 ± 0.51 (t = −1.848; P = 0.107). In MTs, the binocular summation ratio slightly decreased from 1.52 ± 0.54 to 1.39 ± 0.50 (t = 0.919; P = 0.382). Moreover, it remained stable in NTs (from 1.08 ± 0.10 to 1.09 ± 0.09; t = −0.731; P = 0.489). Figure 2D illustrates the changes from baseline in the binocular summation ratio across all six groups of the trained eye at the trained spatial frequency. No significant improvements were detected when comparing trained groups to the controls (ATs vs. ACs, t = −0.126, P = 0.902; MTs vs. MCs, t = 0.207, P = 0.839; NTs vs. NCs, t = −1.992, P = 0.066). 

Perceptual Learning Enhances Visual Functions of the Trained Eye at Untrained Frequencies and in the Untrained Eye

Monocular contrast sensitivity perceptual learning significantly improved contrast sensitivity at untrained frequencies in both the trained and untrained eyes, as well as binocular contrast sensitivity in ATs and MTs. In contrast, NTs exhibited significant gains only in the trained eye at both trained and untrained frequencies, with no notable changes in the untrained eye or in binocular performance. Furthermore, perceptual learning produced substantial enhancements in the balance point for ATs and MTs, whereas the effect in NTs was comparatively modest. 

Monocular Contrast Sensitivity

As shown in Figure 3, in ATs, the AE exhibited significant improvements in contrast sensitivity at untrained spatial frequencies after 10 days of training. A two-way repeated measures ANOVA revealed a significant main effect of time (before training vs. after training) for the untrained frequency of the trained eye, F(1,7) = 12.002, P = 0.010, partial η² = 0.632, and a significant effect of spatial frequency, F(1.643, 11.500) = 83.664, P < 0.001, partial η² = 0.923. The interaction between time and spatial frequency approached significance (F(4,28) = 2.607, P = 0.057, partial η² = 0.271), suggesting that the improvements in contrast sensitivity may vary across these frequencies. The trained eye in the ATs exhibited the greatest improvement in contrast sensitivity at the trained spatial frequency, while showing limited enhancements at spatial frequencies distant from the trained frequency. The untrained eye (FE) also demonstrated significant improvements across all spatial frequencies tested (0.75 – 16 cpd; F(1,7) = 6.808, P = 0.035, partial η² = 0.493, with a significant main effect of spatial frequency, F(1.767, 2.576) = 67.837, P < 0.001, partial η² = 0.906, and a significant interaction, F(2.576, 18.035) = 3.805, P = 0.007, partial η² = 0.352. Binocular contrast sensitivity also improved significantly after training, F(1,7) = 5.680, P = 0.049, partial η² = 0.448, with marked differences across spatial frequencies, F(2.167, 15.170) = 150.689, P < 0.001, partial η² = 0.956. However, no significant interaction effect was observed, F(5,35) = 1.622, P = 0.180, partial η² = 0.188. 

Figure 3.

 

Pre-training and Post-training contrast sensitivity. Contrast sensitivity measured before and after monocular contrast sensitivity perceptual learning at six spatial frequencies (SF = 0.75, 1.50, 3.00, 6.00, 8.00, or 16.00 cpd) for the trained eye, untrained eye, and binocular vision. (A) ATs (red), (B) MTs (blue), and (C) NTs (green). For each group, lighter-colored circles indicate pre-training mean values, whereas darker-colored circles represent post-training means. Error bars denote the standard error of the mean.

In the MT group, the trained eye (nonDE) similarly showed significant improvements at untrained spatial frequencies, F(1,9) = 11.074, P = 0.009, partial η² = 0.552, with a significant effect of spatial frequency, F(4,36) = 172.808, P < 0.001, partial η² = 0.950, and a significant interaction, F(4,36) = 6.012, P < 0.001, partial η² = 0.400. The untrained eye (DE) also exhibited significant gains, F(1,9) = 10.154, P = 0.011, partial η² = 0.530, with spatial frequency effects, F(1.944, 17.499) = 135.100, P < 0.001, partial η² = 0.938, and a significant interaction, F(2.087, 18.786) = 2.677, P = 0.033, partial η² = 0.229. Binocular contrast sensitivity also improved significantly after training, F(1,9) = 6.258, P = 0.034, partial η² = 0.410, with a significant main effect of spatial frequency, F(2.583, 23.245) = 101.533, P < 0.001, partial η² = 0.919, and an interaction effect approaching significance, F(2.360, 21.238) = 3.111, P = 0.017, partial η² = 0.257. For NTs, the trained eye (nonDE) showed significant improvements post-training, F(1,7) = 7.463, P = 0.029, partial η² = 0.516, with a significant spatial frequency effect, F(1.471, 10.296) = 76.919, P < 0.001, partial η² = 0.917, but no significant interaction effect, F(4,28) = 1.482, P = 0.234, partial η² = 0.175. The untrained eye showed no significant overall improvement, F(1,7) = 1.850, P = 0.216, partial η² = 0.209, although a significant effect of spatial frequency was observed, F(2.196, 15.372) = 66.218, P < 0.001, partial η² = 0.904, and a significant interaction between time and frequency, F(1.571, 10.996) = 6.907, P < 0.001, partial η² = 0.497. Binocular contrast sensitivity showed no significant difference between before and after measurement overall tested spatial frequency,(0.75–16 cpd, F(1,7) = 1.257, P = 0.299, partial η² = 0.152, with significant differences across spatial frequencies, F(1.829, 12.800) = 66.744, P < 0.001, partial η² = 0.905, and an interaction effect approaching significance, F(5,35) = 2.813, P = 0.031, partial η² = 0.287. 

In control groups, no significant improvements were observed in either eye (trained eye or untrained eye) or binocularly at any spatial frequency (0.75–16.00 cpd) (Supplementary Fig. S1). For ACs, the AE showed no significant changes across frequencies, F(1,6) = 0.242, P = 0.639, partial η² = 0.039, although a significant main effect of spatial frequency was present, F(1.712, 10.270) = 46.024, P < 0.001, partial η² = 0.885, without a significant interaction effect, F(5,30) = 1.017, P = 0.422, partial η² = 0.145. Similarly, no significant improvements were found in the FE, F (1,6) = 0.152, P = 0.710, partial η² = 0.025, or binocular contrast sensitivity, F(1,6) = 0.054, P = 0.824, partial η² = 0.009, although both showed significant spatial frequency effects without significant interaction. MCs and NCs exhibited similar patterns, with no significant pre-training to post-training changes in either eye or binocularly, MCs, nonDE, F(1,6) = 0.328, P = 0.589, partial η² = 0.052; DE, F(1,6) = 0.223, P = 0.653, partial η² = 0.036; BE: F(1,6) = 0.184, P = 0.682, partial η² = 0.030; NCs, nonDE, F(1,7) = 0.061, P = 0.812, partial η² = 0.009; DE, F(1,7) = 0.116, P = 0.741, partial η² = 0.016; BE, F(1,7) = 0.083, P = 0.781, partial η² = 0.012. Although spatial frequency effects remained significant, MCs, nonDE, F(1.368, 8.211) = 48.224, P < 0.001, partial η² = 0.889; DE, F(1.394, 8.362) = 53.007, P < 0.001, partial η² = 0.898; BE, F(1.429, 8.574) = 58.413, P < 0.001, partial η² = 0.907; NCs, nonDE, F(1.498, 10.484) = 56.089, P < 0.001, partial η² = 0.889; DE, F(1.519, 10.635) = 50.266, P < 0.001, partial η² = 0.877; BE, F(1.612, 11.285) = 61.347, P < 0.001, partial η² = 0.898. No interaction effects reached significance, MCs, nonDE, F(5,30) = 1.134, P = 0.362, partial η² = 0.159; DE, F(5,30) = 0.937, P = 0.472, partial η² = 0.135; BE, F(5,30) = 1.214, P = 0.326, partial η² = 0.168; NCs, nonDE, F(5,35) = 1.263, P = 0.302, partial η² = 0.153; DE, F(5,35) = 1.462, P = 0.225, partial η² = 0.173; BE, F(5,35) = 1.574, P = 0.196, partial η² = 0.184. 

Binocular Functions

Monocular contrast sensitivity perceptual learning significantly enhanced balance point at untrained spatial frequencies (Fig. 4A). In ATs, balance point increased markedly across all tested frequencies: SF1.5 (Z = −2.521, P = 0.012), SF3 (Z = −2.100, P = 0.036), and SF8 (t = −2.409, P = 0.047). Similarly, the MTs demonstrated significant improvements at SF3 (t = −2.636, P = 0.027) and SF8 (Z = −2.701, P = 0.007), though no significant change was observed at SF1.5 (t = −1.514, P = 0.164). In NTs, a significant enhancement in balance point was evident only at SF8 (t = −2.496, P = 0.041), with no significant differences at SF1.5 (t = −1.451, P = 0.190) or SF3 (Z = −0.332, P = 0.750). In contrast, no significant changes in balance point were detected across any spatial frequency in the control groups. Specifically, for ACs: SF1.5 (t = −0.638, P = 0.539), SF3 (t = −1.770, P = 0.111), SF8 (t = 1.098, P = 0.309); for MCs: SF1.5 (t = 0.853, P = 0.422), SF3 (t = −0.343, P = 0.742), SF8 (Z = −0.459, P = 0.646); and for NCs: SF1.5 (t = 1.530, P = 0.170), SF3 (t = −0.616, P = 0.557), SF8 (t = −0.489, P = 0.640). 

Figure 4.

 

Pre-training and post-training measurements of balance point and binocular summation ratio. (A) Balance point (BP) measurements before and after training for ATs (red), MTs (blue), and NTs (green). (B) Balance point measurements for the control groups: ACs (red), MCs (blue), and NCs (green). (C) Binocular summation ratio measurements before and after training for ATs (red), MTs (blue), and NTs (green). (D) Binocular summation ratio measurements for the control groups: ACs (red), MCs (blue), and NCs (green). In (A–D), squares with error bars indicate group means ± SE, and gray dots represent individual participant data. The shaded area highlights the spatial frequency corresponding with the trained frequency (6 cpd).

In contrast with the balance point, the influence of perceptual learning on binocular summation ratio was relatively weak. In ATs, a significant improvement in binocular summation ratio was observed only at SF3 (t = −2.676, P = 0.032), with no significant changes at SF1.5 (t = −0.343, P = 0.742) or SF8 (t = 0.032, P = 0.975). The MTs exhibited no significant changes in binocular summation ratio across all untrained frequencies: SF1.5 (t = 0.869, P = 0.407), SF3 (Z = −0.051, P = 0.959), and SF8 (Z = −0.357, P = 0.721). Likewise, no significant improvements were observed in the NT group (SF1.5: t = −0.521, P = 0.619; SF3: t = 1.040, P = 0.333; SF8: t = 1.470, P = 0.185). Furthermore, none of the control groups exhibited statistically significant changes in binocular summation ratio at any spatial frequency. For ACs: SF1.5 (t = 0.849, P = 0.424), SF3 (t = −1.159, P = 0.285), SF8 (t = −1.376, P = 0.211); for MCs: SF1.5 (t = −1.263, P = 0.238), SF3 (t = 0.753, P = 0.471), SF8 (Z = −0.459, P = 0.646); and for NCs: SF1.5 (t = 0.272, P = 0.793), SF3 (Z = −0.140, P = 0.889), SF8 (t = 2.005, P = 0.085). 

Differences in the Effects of Perceptual Learning on Suprathreshold and Threshold-level Binocular Functions of the Trained Eye

We found that monocular perceptual learning may have different effects on binocular functions at threshold and suprathreshold levels. In particular, both the ATs and MTs exhibited significant post-training improvements in balance point during the suprathreshold binocular task, with these gains transferring to neighboring spatial frequencies. In contrast, no significant training-related changes were observed in the binocular summation ratio during the threshold-level binocular task. 

To further assess the transfer of improvements in binocular functions at trained and untrained frequencies, we conducted linear fits comparing the normalized enhancements in contrast sensitivity, balance point, and binocular summation ratio (Figs. 5A–C). Monocular perceptual learning showed greater effects on balance point compared with binocular summation ratio, with more notable improvements at spatial frequencies near the trained frequency. This effect was consistent across ATs, MTs, and NTs. Additionally, we compared the normalized improvements in contrast sensitivity between the trained and untrained eyes across both trained and untrained spatial frequencies (Figs. 5D–F). In the ATs, the trained eye demonstrated notably greater gains than the untrained eye. By contrast, the MTs showed similar improvements in both eyes. 

Figure 5.

 

Normalized improvements across spatial frequencies. (A–C) Average improvements in contrast sensitivity, balance point, and binocular summation ratio as functions of spatial frequency for ATs (A), MTs (B), and NTs (C). Improvement magnitudes (the difference between post-training and pre-training results) are expressed in decibels (dB) on the y axis, and spatial frequencies are normalized to the trained frequency (log₂(f / f_train)) on the x axis. Shaded regions demarcate the training frequency (6 cpd). Circles denote mean contrast sensitivity values in the trained eye, squares represent balance point, and triangles indicate binocular summation ratios. Error bars depict standard errors, with linear regression fits shown as straight lines. (D–F) display average contrast sensitivity improvements for both trained and untrained eyes across standard spatial frequencies for ATs (D), MTs (E), and NTs (F).

Figure 6.

 

Pre-training and post-training visual acuity measurements. (A–C) Visual acuity results for trained and untrained eyes before (lighter color) and after (darker color) training across ATs, MTs, and NTs. Individual participant measurements are represented by circles, with gray dashed lines connecting pre-training and post-training outcomes for the same participant. Statistical significance is denoted as follows: *** P < 0.001; **0.001 < P < 0.01; *0.01 < P < 0.05; n.s., not significant. (D–F) Correlations between visual acuity improvements and contrast sensitivity changes at the trained frequency (6 cpd) for ATs, MTs, and NTs. Individual participant results are represented by stars, with lighter colors indicating the untrained eye and darker colors indicating the trained eye. Mean values are depicted as squares with standard error bars.

Perceptual Learning Enhances Visual Acuity

In both the AT and MT groups (Figs. 6A–B), monocular contrast sensitivity perceptual learning resulted in significant improvements in visual acuity for both the trained and untrained eyes. Specifically, in the ATs, visual acuity improved significantly from 0.56 ± 0.30 to 0.40 ± 0.31 logMAR in the trained eye (t = 3.897, P = 0.006) and from −0.025 ± 0.10 to −0.10 ± 0.15 logMAR in the untrained eye (Z = −2.049; P = 0.040). Similarly, the MTs showed significant improvements in the trained eye (from −0.050 ± 0.060 to −0.16 ± 0.042 logMAR; Z = −2.842; P = 0.004) and the untrained eye (from −0.040 ± 0.052 to −0.13 ± 0.052 logMAR; Z = −2.913; P = 0.004). In contrast, the NT group (Fig. 6C) demonstrated only marginal improvements, with no statistically significant changes observed in either the trained eye (from −0.028 to −0.041 logMAR; Z = −1.769; P = 0.077) or the untrained eye (from −0.055 ± 0.050 to −0.078 ± 0.065 logMAR; Z = −1.633; P = 0.102). Moreover, no significant correlations were found between visual acuity improvements and contrast sensitivity gains at the trained frequency (6 cpd) in any of the training groups (Figs. 6D–F): for ATs, the trained eye (AE), r = 0.543, P = 0.165; untrained eye (FE), r = −0.151, P = 0.722; for MTs, the trained eye (nonDE), r = 0.465, P = 0.176; untrained eye (DE), r = 0.360, P = 0.307; and for NTs, the trained eye (nonDE), r = −0.198, P = 0.639; untrained eye (DE), r = −0.399, P = 0.328. Additionally, no significant changes in visual acuity were observed in any participants from the control groups (ACs, MCs, and NCs), who received no intervention between the two measurements. 

Discussion

In this study, we explored the impact of monocular contrast sensitivity visual perceptual learning on contrast-related binocular functions in adults with binocular imbalance (AT and MT) as well as in normal participants. We found that monocular contrast sensitivity visual perceptual learning not only enhances the contrast sensitivity in the trained eye at the trained spatial frequency, but also improves balance point (a suprathreshold binocular function measure) at this frequency and transfers to nearby untrained spatial frequencies in participants with binocular imbalance. However, monocular contrast sensitivity perceptual learning did not lead to a significant improvement in the binocular summation ratio (threshold-level measurements). 

Stereopsis is commonly used for assessing binocular vision clinically. However, most clinical stereotests are designed with random dot patterns, which encompass multiple spatial frequencies.^40,41 In this study, we chose balance point and binocular summation ratio as our measures of binocular vision, because they are recognized for their effectiveness in evaluating single-spatial binocular visual function. Specifically, for balance point measurements, we used binocular orientation combination tasks,^17,34 noted for their accuracy in assessing high-frequency binocular combination in amblyopia, though restricted to an upper limit of 8 cpd owing to the constraints of measurement stimuli and interocular suppression of amblyopia. In this task, stimuli were presented dichoptically through head-mounted goggles with a pixels-per-degree value of 41.566, which limited the accurate display of spatial frequencies of >8 cpd. Furthermore, based on our previous studies^17,34 and pilot testing, observer performance was found to be less reliable at higher spatial frequencies. Therefore, the upper spatial frequency limit for this task was set at 8 cpd. We selected 6 cpd as the training frequency to examine the learning effects at this frequency as well as at nearby untrained frequencies (both higher and lower frequencies). After training, our amblyopic participants showed an average increase in contrast sensitivity of 6.27 ± 3.63 dB, which is lower than the increases of 9.8 ± 7.16 dB and 10.7 ± 4.78 dB (the variance results were estimated based on the data from the figures or derived from the standard error) reported in studies by Zhou et al.¹¹ and Huang et al.¹² respectively. This difference likely stems from our use of a lower training frequency (6 cpd) compared with their cutoff spatial frequencies (average of approximately 7.5 cpd); previous studies suggest a direct correlation between the magnitude of perceptual learning and the training frequency: the higher the trained frequency, the greater the observed gains.^42,43 

Moreover, Chen et al.²⁹ found that monocular contrast sensitivity visual perceptual learning highlighted significant improvements in binocular phase combination tasks (suprathreshold interocular balance measurement at 0.293 cpd), whereas Jia et al.³⁰ did not report similar findings (binocular phase combination tasks, 1 cpd). Chen et al.'s study involved participants with a higher degree of imbalance (balance point, 0.14 ± 0.02) compared with Jia's (0.43 ± 0.21). Our study involved participants with varying degrees of binocular imbalance (amblyopia, balance point, across four spatial frequencies: 0.06 ± 0.076 to 0.17 ± 0.15) and participants with less imbalance (anisometropia, 0.58 ± 0.15 to 0.77 ± 0.18). After training, we observed marked improvements in binocular orientation combination tasks at the trained frequency in both amblyopic and anisometropic groups, with these learning effects transferring to nearby spatial frequencies. However, participants with greater imbalances, such as those with amblyopia, demonstrated the ability to transfer learning effects to a broader range, whereas participants with anisometropia showed no significant learning effects at 1.5 cpd. This finding suggests that greater binocular imbalance may have greater potential and thus enhance the gains from perceptual learning and its transferability to nearby untrained frequencies. 

Furthermore, in our study, the transferability of learning effects in monocular contrast sensitivity perceptual learning was more pronounced at suprathreshold levels. Huang et al.²⁸ proposed a multipathway contrast-gain control model, suggesting that changes in monocular signals might affect interocular contrast-gain control, which in turn changes the weights of two eyes in binocular combination. Additionally, the model indicates that, although the computation of different features (e.g., phase and contrast) first shares the same interocular contrast-gain controls, the subsequent pathways for feature specific calculation could be different. Our findings demonstrate that, although perceptual learning had transfer effects on both balance point and binocular summation ratio, the impact on binocular summation ratio was notably weaker. This observation is consistent with the framework of the multipathway contrast-gain control model. 

The prevailing consensus in current studies is that amblyopia primarily results from changes in the neural circuits of V1, driven by alterations in visual experience and enhanced neural plasticity during cortical development.^44,45 In models of amblyopic macaques and cats, some V1 neurons have shown altered spatial frequency tuning and a loss of contrast sensitivity when receiving inputs from the AE, and the proportion of binocularly activated cells in the V1 area of some amblyopic monkeys has significantly decreased, along with a decrease in the proportion of neurons responsive to the AE stimulus.^46–48 Deficits in contrast processing associated with amblyopia can occur both downstream of V1 and at the level of neuronal correlations within V1.⁴⁹ Perceptual learning has been observed to increase contrast gain in V1 neurons. After undergoing contrast sensitivity visual perceptual learning, Hua et al.¹³ noted a significant enhancement in the contrast sensitivity of V1 neurons in cats, with preferred spatial frequencies close to those trained, and this induced plasticity was attributed to increased contrast gain in training-related neurons. In adults, perceptual learning selectively increased functional magnetic resonance imaging responses to low-contrast patterns in the magnocellular LGN, with no changes in the parvocellular layers or cortex. This enhancement correlated with behavioral gains, suggesting thalamic-level plasticity in LGN neurons.¹⁴ Furthermore, perceptual learning has been reported to strengthen synaptic connections within V1,⁵⁰ inducing long-term potentiation of synaptic responses in rat V1⁵¹ and enhancing the feedforward connections from the LGN to V1.⁵² In macaque models, the majority of cells in the primary visual cortex are binocularly innervated,^53,54 suggesting that the brain is a true binocular system, even under monocular observation conditions, where the process necessarily involves downstream cortical areas sensitive to binocular input. Contrast sensitivity serves as a fundamental function reflecting the output of early visual processing, representing the performance of primary visual cortex neurons, and its enhancement could potentially offer significant benefits for later stages of visual processing. 

Given recent revelations that adults still retain some potential for visual plasticity, increasingly diverse treatments for amblyopia, especially in adults, have been developed.⁵ Although the definition and diagnosis of amblyopia focus on the impaired best-corrected visual acuity of the AE, traditional treatments primarily aim to improve this aspect. However, as the understanding of amblyopia as a binocular disorder has evolved, more treatment modalities are being explored and discussed for their potential benefits to binocular functions in amblyopia.⁵⁵ Notably, an essential consideration in clinical applications is the cost-effectiveness of the treatment method, that is, whether a simple approach can yield more significant benefits. Our study indicates that simple contrast detection tasks used in perceptual learning not only enhance monocular functions at the trained frequency, but also positively affect binocular functions at the trained frequency and nearby untrained frequencies. This demonstrates the transfer characteristics of monocular contrast perceptual learning to contrast-related binocular tasks and underscores the clinical application value of this treatment method. 

To evaluate test–retest reliability, we included untrained control groups who completed only test and retest sessions without undergoing monocular contrast sensitivity perceptual learning. These control participants showed no significant improvements in contrast sensitivity, balance point, or binocular summation ratio, nor did their results exhibit spatial frequency specificity. This work highlights the specificity of the training-induced enhancements observed in the trained groups. Consistent with our findings, previous studies comparing perceptual learning groups with untrained controls have similarly reported significant contrast sensitivity improvements exclusively in the trained groups, with no notable changes in the controls.⁵⁶ Together, these results confirm that the observed improvements stem from perceptual learning rather than from repeated testing alone. 

To minimize the interference of previous training experiences on current training outcomes, all participants in our study had no history of perceptual learning. However, this added complexity to participant recruitment, resulting in a smaller sample size. Additionally, constrained by the measurement tasks and measurement duration, we did not assess participants' binocular visual functions across a broader spectrum of spatial frequencies. We hope that with advancements in measurement technologies, future studies will be able to conduct more comprehensive investigations into the characteristics and mechanisms of perceptual learning in patients with binocular imbalances. 

Acknowledgments

The authors thank all subjects for participating in the study. 

Supported by the National Key Research and Development Program of China Grant (2023YFC3604104), the National Natural Science Foundation of China grant (82471118), and the Non-profit Central Research Institute Fund of Chinese Academy of Medical Sciences (2023-PT320-04) to J. Z., and the National Natural Science Foundation of China Grant (NSFC U23A20437) to Z.H. 

Disclosure: W. Lin, None; Z. He, None; S. Zhou, None; L. Weng, None; L. Zou, None; R. Ye, None; J. Zhu, None; J. Zhou, None