Perceptual learning improves visual functions in patients with albinistic bilateral amblyopia: A pilot study

1 Introduction

Albinism in humans is a congenital disorder discriminable by a partial or complete absence of the biosynthesis of melanin, causing hypopigmentation of the eyes, hair and skin (Scheinfeld, 2003). The reduction of melanin, which contributes to the development of the visual system, may lead to changes in the visual system such as misrouting of the retinogeniculate projections, foveal hypoplasia and alterations in the cortical sensory representation (Ather et al., 2019; Bridge et al., 2014; Gottlob, 2007). Associated to albinism, several functional visual conditions may be present such as ametropia, astigmatism, nystagmus, photophobia 1 . Bilateral amblyopia can also develop in association with this organic ocular disease (Holmes & Clarke, 2006). Differently from unilateral amblyopia 2 , that arises after a selective inability of the brain to process inputs from one eye, in bilateral amblyopia the brain fails to process the information coming from both eyes. In both cases, the visual impairment does not improve with optical corrections. One major issue is whether albinistic bilateral amblyopia (ABA) can be treated with positive outcomes, using the same protocols found to be successful in the treatment of unilateral functional amblyopia.

The most common initial treatment for functional amblyopia, not associated to organic cause, is the occlusion therapy: by patching the fellow eye, the activation of the amblyopic eye is boosted (Wallace, 2006; Wallace et al., 2013). Although patching improves visual acuity, this method is not uncontested, since it is difficult to comply with, it leaves residual deficits in the amblyopic eye and, most importantly, it may lead to occlusion amblyopia in the fellow eye (Polat, Ma-Naim, Belkin, & Sagi, 2004; Varadharajan & Hussaindeen, 2012). Extensive research has provided evidence for alternative treatment options. These include, monocular perceptual learning (Barollo, Contemori, Battaglini, Pavan, & Casco, 2017; Li, Provost, & Levi, 2007; Li, Young, Hoenig, & Levi, 2005; Polat, 2008) and dichoptic treatment protocols (Hamm et al., 2018; Knox, Simmers, Gray, & Cleary, 2012; Li et al., 2013; Vedamurthy, Nahum, Bavelier, & Levi, 2015).

There are a few published studies on treatment for bilateral amblyopia (Haase, 1978; Hamm et al., 2018; Jeon, Maurer, & Lewis, 2012; Schoenleber & Crouch, 1987; Taylor, Powell, Hatt, & Stewart, 2012; Wallace et al., 2007) and to our knowledge none of them had focused on bilateral amblyopia due to albinism. Most commonly, the treatment of bilateral amblyopia is done with early refractive correction. Patching has been rarely used in bilateral amblyopic patients with sustained unequal acuity between the two eyes (Taylor et al., 2012). Few studies have assessed the effect of alternative techniques, such as video game training (Jeon et al., 2012) and contrast-balanced binocular treatment (Hamm et al., 2018).

In the present study, we aimed at exploring the efficacy of a training protocol based on perceptual learning (PL), in improving visual functions in patients with organic bilateral amblyopia due to albinism. There is extensive evidence that PL is a powerful technique to restore visual functions in patients with functional amblyopia. PL consists in a performance improvement obtained after repeated practice in a visual task, associated with neural plasticity outside the critical period (see Sagi, 2011 for a review). Treatment of monocular functional amblyopia through PL has been carried out using several training tasks: identification of letters (Chung, Li, & Levi, 2006, 2008; Levi, 2005); discrimination of relative position of line segments (Levi & Polat, 1996; Levi, Polat, & Hu, 1997) and detection of contrast defined stimuli either isolated (Huang, Lu, & Zhou, 2009; Huang, Zhou, & Lu, 2008; Zhou et al., 2006) or presented together with collinear flankers in a lateral masking configuration (Barollo et al., 2017; Chen & Tyler, 2008; Polat, et al., 2004; Polat, Ma-Naim, & Spierer, 2009), or even associated to neurostimulation (Campana, Camilleri, Pavan, Veronese, & Lo Giudice, 2014).

Although the aforementioned tasks have been shown to produce PL-dependent improvement in visual acuity for letters, the contrast detection task of a Gabor patch performed in a lateral masking regimen 3 seems to produce a greater and long-lasting improvement, as well as a higher level of generalization to several untrained tasks, such as Vernier acuity, contrast sensitivity and crowding (Levi, 2012). Moreover, it seems more suitable to pinpoint the neural level at which the improvement occurs. Indeed, previous studies suggest that training the contrast-detection task of a Gabor patch performed in a lateral masking regimen produces training-induced cortical plasticity (Maniglia, Soler, Cottereau, & Trotter, 2018), outside the critical period (Sagi, 2011), at the level of the early visual cortices (Gilbert & Wiesel, 1985; Grinvald, Lieke, Frostig, & Hildesheim, 1994; Ts’o, Gilbert, & Wiesel, 1986). This task is mediated by the activation of excitatory (E) and inhibitory (I) subpopulations of neurons in a cortical column of the primary visual cortex, with the ratio between E and I activation increasing as a consequence of two inputs: target contrast (thalamic input) and the input from lateral cortical connections biased versus excitation (Adini, Sagi, & Tsodyks, 1997; Battaglini et al., 2019; Chen, Kasamatsu, Polat, & Norcia, 2001; Polat, 1999; Polat, Mizobe, Pettet, Kasamatsu, & Norcia, 1998; Seriès, Lorenceau, & Frégnac, 2003; Stemmler, Usher, & Niebur, 1995). In particular, this last input can be modulated by lateral masking producing threshold elevation (E/I < 1) when flankers are presented close to the foveal target (< 2λ) and threshold reduction (E/I > 1) when they are presented at a distance between 3–6λ, with λ indicating the wavelength. Considering the alterations of the neural representation of the sensory input in ABA (Ather et al., 2019; Bridge et al., 2014; Gottlob, 2007), the employment of PL using a lateral masking paradigm may help to induce positive plasticity 4 in the early visual system of albinos. If perceptual learning occurs, then two scenarios (not mutually exclusive) will be possible. PL will result either in a reduction of neural noise with a consequent increase in contrast gain for the target, or it will also produce an effect consisting of the modulation of collinear lateral interactions (excitation and/or inhibition exerted by the flankers), at the early cortical level, (Adini, Sagi, & Tsodyks, 2002; Battaglini et al., 2019; Lev & Polat, 2015; Seriès et al., 2003; Stemmler et al., 1995).

Six patients with oculocutaneous albinism (OCA) and ABA, defined as 0.2 logMAR acuity or worse, poor or absent stereoacuity and a diagnosis of albinism, and six normal sighted controls, participated in a PL protocol, based on a contrast-detection task with lateral masking. The task was to choose in which of the two consecutive time intervals a low-contrast Gabor patch was presented centrally in the fovea. The target was presented with two flankers with the same spatial frequency but a higher contrast. We measured how the contrast threshold for the target was modulated by target-to-flanker separations, which is the parameter that allows switching lateral interactions from inhibitory to excitatory, at short and medium distances, respectively. We inferred neural plasticity at the cortical level if, in addition to a reduction of the contrast threshold for the target after training, either inhibition or excitation by the flankers was modulated by PL. We also probed the generalization effects of the PL to higher-level visual functions (visual acuity, contrast sensitivity function, hyperacuity, and foveal crowding).

2 Methods

2.3 Stimuli

2.3.1 Perceptual learning stimuli

PL stimuli were generated using Matlab Psychtoolbox (Brainard, 1997; Pelli, 1997). The software used to generate the stimuli was set up for the screen resolution to ensure the correct representation of spatial frequencies at the viewing distance of 57 cm.

The stimuli were configurations of either two or three Gabor patches, each consisting of a cosinusoidal carrier enveloped by a stationary Gaussian, according to the following formula: G ( x , y ) = cos ( 2 π λ x + φ ) e ( - x 2 + y 2 σ 2 )(1)

Each Gabor patch was characterized by its wavelength (λ), its phase (φ) and the standard deviation (σ) of the luminance Gaussian envelope in the space image (x, y). σ=λ and φ= 0 (even symmetric). Gabors’ spatial frequencies used during the training ranged between 1 and 15 cycles/deg (c/°). The target was the central Gabor. It was presented flanked by two collinear high contrast Gabor patches (Michelson contrast = 0.7) with the same orientation and spatial frequency as the ones of the target (Barollo et al., 2017). Four orientations were possible: horizontal, vertical, diagonal (45° and 135°). Stimulus-to-flankers distances were 2λ, 3λ, 4λ and 8λ, as shown in Fig. 1. Training transfer tasks, described below, were performed in a dark room. Participants wore their glasses or contact lenses when required.

OPEN IN VIEWER

Fig. 1

Vertical configuration; from left to right: 2λ, 3λ, 4λ and 8λ.

2.3.2 Transfer stimuli

Before and after the training, far and near visual acuity, contrast sensitivity function (CSF), foveal crowding and Vernier acuity were measured at the clinical centre. Far Visual acuity was tested both monocularly (both left and right eye) and binocularly. Viewing was binocular in the other tests.

2.3.2.1. Far visual acuity stimuli: Font size threshold for far visual acuity (hereafter farVA) was measured using the Freiburg Visual Acuity and Contrast Test (FrACT) software (Bach, 1996), that uses as a target stimuli to identify 10 Sloan letters C, D, H, K, N, O, R, S, V and Z. Participants were asked to identify and report the presented letters. The responses were typed by the experimenter. Any target Sloan letter misidentified as a different target letter or a non-target letter was recorded as an error.

2.3.2.2. Near visual acuity and crowding: Foveally presented stimuli to measure near visual acuity (hereafter nearVA) and crowding were generated using Matlab Psychtoolbox (Brainard, 1997; Pelli, 1997). First, the threshold for near visual acuity was assessed by presenting a set of Sloan letters at 0 deg eccentricity. Letters were white on a black background. Once threshold for near central visual acuity had been measured, foveal crowding was assessed. The target stimulus was a Sloan letter flanked vertically by two random Sloan letters. We measured the minimum target-to-flanker distance at which the flankers increased target threshold. In both tasks, participants read aloud the target letter which was then typed by the experimenter.

2.3.2.3. Contrast sensitivity function (CSF) stimuli: CSF was tested using the FrACT software (Bach, 1996). Participants had to report to the experimenter the orientation (horizontal, vertical, diagonal at 45°, and 135°) of a grating stimulus of 2° in diameter viewed within a circular window. Spatial frequencies tested were 1, 3, 5, and 7 c/°.

2.3.2.4. Vernier acuity: Vernier acuity was tested using the FrACT (Bach, 1996) software. The stimulus was a two-line vertical segment configuration (15 × 1.4 arcmin) with a vertical separation between the lines of 0.5 arcsec. The task was to discriminate whether the upper segment was displaced to the left or the right to the lower segment. The answer was typed by the experimenter.

This work aimed to probe if bilateral organic albinistic amblyopia due to OCA can be treated. More specifically, the question addressed was if ABA can be treated using the same PL protocol used for functional amblyopia (Barollo et al., 2017; Chen & Tyler, 2008; Polat, et al., 2004; Polat, et al., 2009).

Data indicate an effect of training in the ABA group at 2λ and 3λ. The control group did not seem to have a positive training effect. This was expected considering that the training occurred in fovea, where contrast sensitivity is already very high (Sagi, 2011; cf. Polat & Sagi, 1994) even though, in previous studies (Adini et al., 2002; Casco et al., 2014) the duration of the training of the control sample was shown to be long enough to produce PL. For what is concerning the untrained transfer tasks, we found a general improvement in the ABA group, but results were not consistent among tasks. In farVA with Sloan letters, after Bonferroni correction, the improvement was observed only in the RE, where we measured higher visual acuity in the pre-test in four out of six ABA patients. These results are apparently in conflict with a recent study (Moshkovitz et al., 2020), that showed an improvement in both eyes after training in albinism. However, there are some differences between Moshkovitz and colleagues (2020) and our study: i) we have a smaller sample size and less statistical power, ii) in the sample collected by Moshkovitz et al., (2020) there were also patients with infantile nystagmus, without oculocutaneous albinism and finally iii) they did not test the improvement in VA after training. Therefore, a direct comparison between the two studies is not truly possible. Contrast sensitivity measured with an orientation discrimination task (FrACT) also improved in ABA group with a Gabor patch of 3 c/°. Data from the crowding task showed significant improvement in the ABA group. On the other hand, results observed in NearVA and Vernier tasks did not reach significance.

To summarize, the effect of PL on visual acuity is consistent with previous studies on non-albino amblyopic subjects (see Barollo et al., 2017; Polat et al., 2004). Similarly, for the contrast sensitivity function, that was measured with an orientation discrimination task rather than a detection task, a larger effect of the training was found in the ABA group, but only with a Gabor target of 3 c/°. Moreover, a promising effect of reduced crowding after PL was observed.

Besides the transfer effects, we were interested to know whether the training effect was only due to contrast gain for the target, or it also reflected a modulation of lateral interactions between target and flankers, at the early cortical level. To address the effect of lateral interactions we normalized the thresholds obtained at 2-3 and 4λ, by those obtained in the baseline 8λ condition, where the flankers should not affect the contrast threshold for the target. The lateral masking paradigm has been often used in normal vision to evaluate the role of lateral interactions on the contrast threshold for the target both in normally sighted and unilateral amblyopic participants. In the fovea, facilitation by the flankers for a range of 3-4λ was consistently found in previous studies (Polat, Bonneh, Ma-Naim, Belkin, & Sagi, 2005; Polat et al., 2004; Polat, Sagi, & Norcia, 1997). At 2λ some studies found facilitation (Polat et al., 2005; Polat et al., 2004), some other found inhibition (Lev & Polat, 2011, 2015; Polat & Norcia, 1996; Polat et al., 1997), whereas we did not find a clear effect of the flankers. For target-to-flanker distance >than 6λ, there is agreement that lateral interactions are absent when presentation is central (Polat et al., 2005) and they are present when it is peripheral (Lev & Polat, 2011, 2015; Maniglia et al., 2011). Overall, our control group performance is in good agreement with what was found in previous studies. ABA group did not show any lateral masking effect before the training. This result, in particular the absence of inhibition, is similar to what found for unilateral functional amblyiopia due to strabism (Bonneh, Sagi, & Polat, 2004; Polat et al., 2005; Polat et al., 1997), whereas anisometropic patients showed facilitation in particular at 3 and 4λ (Polat et al., 2005; Polat et al., 2004; Polat & Norcia, 1996). Therefore, the type of amblyopia does play a role in modulating lateral interactions in the brain. However, we feel that we cannot go further in associating the ABA to one or the other type of unilateral amblyopia, since anomalous spatial interactions depend on which parameters have been combined, namely spatial frequency, amblyopia type and astigmatic axis (Polat et al., 2005). We find a three-way interaction group×  training×λ that suggests an effect of PL on the lateral interactions. For normally sighted observers, modulation of lateral interactions is not likely to be expected when stimuli are centrally presented (Sagi, 2011; cf. Polat & Sagi, 1994). Indeed, we did not find any training effect in this group that underwent a training duration comparable to that of previous studies on PL with lateral masking paradigm in the periphery (Shani & Sagi, 2005), although the duration was shorter than that of the ABA group. Concerning unilateral functional amblyopia, the effect of PL is not clear. Polat et al (2004) found that training with a target of 6 c/° reduced pre-training inhibition at 3λ. Barollo et al. (2017) found a reduction of threshold as a consequence of training with a 3 c/° target, stronger at 8λ than a 3 and 4λ. In the present study trained target had spatial frequencies ≥ 5 c/° (except P1 that was trained at 3 c/°) and the effect of training on lateral interactions is consistent with what was found by Polat et al. (2004) (reduced pre-training inhibition at 3λ). To sum up, our results indicate that patients with bilateral organic amblyopia have a similar advantage from flankers as patients with functional unilateral amblyopia when target spatial frequency is approximatively matched. This conclusion needs however to be confirmed in further work with a larger sample size.

In the attempt of explaining why the difference in amblyopia type, either organic or functional, makes no substantial difference with respect to the increase of facilitatory lateral interactions following PL, two interpretations may be put forward. One relies on the evidence that the amount and polarity of lateral masking in contrast detection rely on the subpopulation of neurons in a cortical column, having either excitatory (E) or inhibitory (I) response. Increasing the contrast for the target (thalamic input) would make excitation increase more than inhibition (Lev & Polat, 2015). On the other hand, for target contrast lower than that of the flankers, lateral input depends on target flanker separation and target contrast. It would be biased towards excitation when target and flankers separation ranges between 3 and 4λ and contrast is relatively low, whereas inhibition would prevail if target flanker separation is 2λ or smaller and the target contrast is relatively high (Adini et al., 2002; Battaglini et al., 2019; Chen et al., 2001; Seriès et al., 2003; Stemmler et al., 1995). The hypothesis that this mechanism may be modulated by PL both in organic and functional amblyopia implies that lateral input at early cortical level plays a role in both types of amblyopia. Moreover, it should be considered that the effect of lateral interactions results from a complex interaction between the contrast of the target and that of the flankers at each target to flanker distance (Zenger & Sagi, 1996). To be brought down to a maximum simplification, the model of Zenger and Sagi predicts that for a target-to flanker separation of 3-4λ, facilitation by the flankers would occur if target contrast is about 0.25 of flanker contrast. Conversely, as the flanker/target contrast approaches 1 (i.e. the contrast threshold for the target increases), inhibition would prevail at the same target-to-flanker distance of 3λ. Therefore, considering that target contrast at threshold for the ABA group decreases from about 0.5 to 0.2 of flankers contrast after training, this shift could well account for the shift between the absence of the effect and the facilitatory effect by the flankers after PL.

Another explanation is that for both types of amblyopia PL may enhance the capability to efficiently extract the information from the target, possibly improving the efficiency of neural decision mechanisms at the highest level of central processing (Levi, 2012; Li, Klein, & Levi, 2008). In other words, as a result of PL, the amblyopic visual system might improve the capability of responding to the target input by reducing internal noise through a mechanism relying on a more synchronized neural network (Scheler, 2018). Moreover, the involvement of a high-level neural decisional mechanism in PL may explain the substantial transfer effects that usually accompany PL. Other learning effects, such as learning to fixate and/or accommodate more accurately are very unlikely. Indeed, if the improvement was accounted for by improved fixation or focus, we would expect, in the CSF an improvement at the range of spatial frequency trained, although the task changes from detection (during PL) to orientation discrimination (in the CSF). Instead, we only found transfer to untrained lower spatial frequencies.

We acknowledge that this study has some limitations. The sample size was very small and further studies with a larger number of participants are needed to confirm our findings. Moreover, the mean age of the control group is a bit higher than the mean age of the ABA group (t(5) = 4.14; p = 0.053). However, we believe that the results of the current study are important for guiding future investigation in this field.

In conclusion, we showed that humans suffering from organic amblyopia obtained a high degree of neural modulation from PL which is not obtained after a similar training regime in normal foveal vision.

These results have important clinical applications because they open new opportunities for the rehabilitation of children and adults with organic amblyopia that cannot take advantage of the typical rehabilitative techniques, such as patching and refractive correction.

Disclosure of funding sources

The study was supported by a grant from MIUR (Dipartimenti di Eccellenza DM 11/05/2017 n.262) to the Department of General Psychology

Declaration of competing interest

The authors declare that they have no conflict of interest related to publication of this manuscript.