The Distribution of Spatial Contrast in Manmade Environments Produces Regional Visual Form Deprivation

Abstract

Purpose

Children who spend less time outdoors or who live in urban areas are more likely to develop myopia (short-sightedness). This may stem from the altered spatial distribution of contrast in manmade environments, which contain large featureless surfaces that may potentially produce regional environmental form deprivation (eFD).

Methods

Images (n = 590) from natural, mixed-urban, urban, and indoor environments, were subdivided into 36 × 36 zones (1.6° × 1.2° visual angle). The weighted contrast energy (CE_w) of each zone was calculated by filtering the fast Fourier transform with a filter describing the contrast sensitivity of the human retina at that eccentricity. Zones with less CE_w than in images taken through white Perspex ocular diffusers that induce form-deprivation myopia in guinea pigs were classified as eFD. The spatial complexity of CE_w signals across the visual field were compared among the environments.

Results

Featureless manmade structures such as blank walls, ceilings, and roads contain low CE_w and thus cause regional eFD, particularly when present in the peripheral retina where spatial resolution is low. Based on the images analyzed from indoor environments, 29.5% of the human visual field would potentially experience eFD, a percentage greater than that likely experienced in outdoor urban (15.2%, P < 0.001), mixed-urban (6.7%, P < 0.001), and natural (2.8%, P < 0.001) environments. The complexity of the contrast energies perceived across the visual field were also significantly less in indoor environments than in all other settings, meaning eye movements are less likely to result in a change in contrast, thus regional eFD is more likely to be temporally maintained. Furthermore, this effect was exaggerated in the visual periphery during gaze-limiting tasks such as reading or using a mobile phone.

Conclusions

Widespread eFD is potentially experienced across the visual field in manmade environments. The extensive peripheral eFD and diminished contrast complexity of manmade environments suggest that gaze-limiting activities such as reading or using a mobile phone result in extended periods of peripheral eFD, which may contribute to myopia development. Therefore, enhancing the distribution of contrast in artificial environments to limit regional eFD may serve to prevent myopia onset.

Myopia (short-sightedness) is a refractive disorder whereby the length of the eye exceeds the power of the anterior optics. The prevalence of myopia has exploded in recent decades, particularly among younger generations, with as many as 80% of school leavers^1,2 and 95% of university students³ displaying myopia in regions of east and southeast Asia. However, the myopia boom is not limited to Asia. Based on population modeling from 2016, approximately half of the global population is predicted to be myopic by 2050.⁴ This increase is too rapid to be genetic in origin, suggesting some factor of our modern lifestyle is driving such a change.

Animal studies have demonstrated an evolutionarily conserved link between visual environment and the regulation of axial eye growth during adolescence. This is highly relevant to the current human myopia epidemic, as the greatest environmental risk factors for myopia are increased educational pressure,^5,6 total duration of education,^2,7 reduced time outdoors,^8–13 and urbanicity of environment,^14–18 each of which exemplifies a clear deviation from a natural visual environment.

A sense of how such environments could promote myopia development in humans can be gained from considering visual manipulations that cause myopia in animal models. For example, across species, the rate of axial elongation is regulated by the retina to counteract the sign and power of imposed optical defocus,^19–24 precisely repositioning the retina to the focal plane of the anterior optics. This response likely underpins the process of emmetropization in humans, as the eye begins as hyperopic and gradually aligns the retina with the focal plane of the anterior optics. Furthermore, this phenomenon underlies the proposed mechanism through which myopia treatment lenses containing positive defocus slow myopia progression.^25–28

Myopia may also be induced by form deprivation (FD), where wearing a diffuser over one eye that deprives the eye of detailed vision (thus interrupting the normal spatial frequency and contrast range exposure) results in accelerated axial elongation and myopia formation.^29–33 Here, FD is used to specifically refer to the visual stimulus that is associated with the induction of form-deprivation myopia (FDM) in animal models. FDM is an open-loop response; as unlike signals arising from optical defocus, the contrast attenuation from the diffuser is not rectified by axial elongation. Also, FDM is a graded phenomenon, as in guinea pigs,³⁴ chicks,³⁵ and rhesus monkeys,³⁶ Bangerter foils induce ocular elongation and myopia, the severity of which increases with the extent of contrast attenuation of the foil. As such, differences in the contrast composition between urban and rural environments and between indoor and outdoor environments could help to explain the associated increased risk of myopia development via the same mechanisms underlying FDM in animal models.

With respect to contrast, images of natural outdoor scenes display common statistical characteristics when transformed to frequency space. For example, radially averaged amplitude spectra (reflecting the amplitude, or difference in luminous intensity averaged across all orientations) decay with spatial frequency (f) at a rate of approximately 1/f,^37,38 meaning that the relative contrast energy/f of a natural scene is generally independent of image scale or viewing distance.³⁸ This scale invariance is disrupted in environments laden with feature-poor manmade structures such as in outdoor environments in urban areas and indoor environments, which exhibit steeper spectral amplitude-frequency slopes of 1/1.24f and 1/1.50f, respectively (weighted averages from all images).³⁷ On average, this means that outdoor urban scenes and indoor environments contain comparatively less contrast energy at middle and higher frequencies than in nature and that, depending on the composition of the environment, this deficit may increase further with reduced viewing distance.

A deficit in adequate contrast exposure may contribute to the higher prevalence of myopia in modern humans, as rearing chicks in artificial environments with steeper spectral amplitude–frequency slopes of 1/2f induces substantial axial elongation and myopia relative to chicks reared in environments with slopes of 1/f and 1/0.5f.³⁹ Furthermore, the spectral amplitude slope of images taken of children's bedrooms and study areas in a Chinese population were significantly associated with refractive error.⁴⁰ The steeper average spectral amplitude gradients of indoor environments can be recreated by imaging a natural scene through a Bangerter foil, suggesting that these visual environments may elicit similar adaptive responses. However, the mean spectral amplitude slope is a simplified measure that reflects the average contrast experienced across the visual field and ignores any regional differences in the spatial distribution of contrast experience by the eye. In the current study, we sought to determine the impact of these regional differences perceived across the retina.

Animal studies have demonstrated that myopiagenic stimuli such as FD are interpreted regionally by the retina, producing regional elongation.^41–44 Conceivably, regions of the human retina observing feature-poor surfaces in manmade environments entirely devoid of contrast/form would thus experience the strongest stimuli for elongation. This threshold of FD would also be dependent on the contrast sensitivity of the human retina at that eccentricity, which declines toward the visual periphery due to reduced retinal cell density.⁴⁵ Thus, the human visual periphery would be more likely to experience strong FD stimuli than the central retina in manmade environments. In primates,⁴⁶ FD in the visual periphery causes significant axial elongation along the central visual axis and myopia formation,⁴⁶ suggesting that, if the peripheral human retina experiences sufficient FD, this could contribute to myopia formation. In the present study, it was hypothesized that the presence of feature- and contrast-poor structures in manmade environments causes regional visual stimuli that mimic FD, which we term environmental form deprivation (eFD), in the visual periphery which may contribute to myopia development.

To investigate this, the regional contrast energy in a series of images from manmade and natural environments was calculated for spatial frequencies resolvable at each visual eccentricity by the human retina. These maps were compared to those from images taken through solid white plastic ocular diffusers which entirely attenuate contrast at all but the lowest spatial frequencies and induce the maximum rate of myopia development in our mammalian model of myopia,^33,34 allowing regions of the retina likely exposed to visual environments associated with myopia development to be mapped for different environments. Here, we show that the same visual cues that cause myopia development in animal models are experienced across large areas of the human visual field in manmade environments. Specifically, blank, featureless structures in urban and indoor environments result in extensive pockets of visual eFD, most prevalent in the visual periphery, a region important for driving changes in eye growth and myopia formation.^41,47 The diminished complexity of contrast signals across the visual field in manmade environments means that regional eFD is more likely to be maintained even after eye movements. Furthermore, we describe how behaviors that limit eye movement such as reading or using a mobile phone could produce prolonged deprivation, particularly in the visual periphery, due to the diminished complexity of regional contrast signals across the visual field in indoor environments. Finally, we show that images filtered through a Bangerter foil, which also induces myopia in animals, similarly increases the amount of eFD perceived across the visual field in all environments due to the attenuation of contrast.

Methods

Image Acquisition and Processing

Digital color images (n = 590; resolution, 4032 × 3024 pixels) were captured opportunistically at eye level across Singapore using an iPhone SE (2nd generation; Apple, Cupertino, CA, USA), over the course of 6 months. Images were classified into four categories based on environmental composition: indoor (n = 131); urban (n = 137), which included manmade outdoor environments with no natural greenery (pot plants excluded); mixed urban (n = 171), which included outdoor environments integrating natural and manmade elements; and natural (n = 151), taken in national parks, on nature walks, and in community gardens. Examples from each category are included in Supplementary Figure S1. Images were converted to 8-bit grayscale images using the rgb2gray function in MATLAB (MathWorks, Natick, MA, USA) (Fig. 1a) and divided into 36 × 36 rectangular zones of equal size, each equating to 1.66 × 1.25 degrees visual angle (dva), based on the original image occupying 60 × 45 dva (Fig. 1b). For each zone, the two-dimensional (2D) amplitude spectra (X[k]) (Fig. 1c) were calculated using the 2D fast Fourier transform in MATLAB.

Figure 1.

Image processing pipeline to determine regional eFD. (a) An example image from a mixed-urban environment (containing manmade and natural components), following grayscale conversion. (b) The image was segregated into 36 × 36 zones, and each zone was classified based on its eccentricity (visual angle). (c) Example fast Fourier transform of a single image zone. (d) Two-dimensional doughnut filter equating to the human contrast sensitivity function applied to the transformed zone. (e) Filtered magnitude spectrum of an image zone. (f) The equation used to quantify the weighted contrast energy in each filtered transform. (g) An example heat map displaying the weighted contrast energy (CE_w) of each image zone from the example image. (h) The original grayscale image with regions of eFD highlighted in red.

Calculating the Weighted-Contrast Energy in Each Image Zone

The total contrast energy at all spatial frequencies and orientations can be calculated using Equation 1:

$$\begin{array}{c}\hfill E_{contrast}=\frac{1}{N}\sum _{n=0}^{N-1}{\left|X\left[k\right]\right|}^2\end{array}$$ (1)

That is, by taking the sum of the squared absolute values of the amplitude spectra (|X[k]|) divided by the number of sample points (N, number of pixels) of each transform. However, due to variation in resolution across the human retina, simply equating the total contrast energy of the image would overestimate the degree of contrast resolvable by the retina, particularly for the retinal periphery. Therefore, the total contrast energy contained within each image zone (within the resolvable spatial frequency range of the corresponding region of retina) was calculated. To achieve this, each 2D amplitude spectrum was multiplied by a log-Gaussian “doughnut” filter, L_gaus(f) (see Equation 2, Fig. 1d):

$$\begin{array}{c}\hfill L_{gaus}\left(f\right)=\mathrm{ex}{\mathrm{p}}^{\left\{-\left[\frac{ln{\left(\frac{f}{f_{peak}}\right)}^2}{2ln\left(\frac{f_{\sigma }}{f_{peak}}\right)}\right]\right\}}\end{array}$$ (2)

where f refers to radial frequency $f=\sqrt{\left(f_x^2+f_y^2\right)}$, f_peak is the spatial frequency with peak contrast sensitivity and f_σ is the bandwidth of the filter. The values for f_peak and f_σ were determined by fitting the psychophysical measurements of achromatic contrast sensitivity in human subjects at 0°, 5°, 10°, 15°, 20°, and 27° visual angle, previously published by Chwesiuk et al.,⁴⁸ using Equation 3 (Fig. 2), which multiplies the log-Gaussian function (Equation 2) by the peak contrast sensitivity (CS_max) in that region:

$$\begin{array}{c}\hfill CSF=C{S}_{max}*L_{gaus}\left(f\right)\end{array}$$ (3)

Figure 2.

Relative contrast sensitivities at different visual eccentricities as reported by Chwesiuk et al.⁴⁸ The log-Gaussian fits were applied to mean achromatic contrast sensitivity data at 0, 5, 10, 15, 20, and 27 dva from five participants. The right y-axis details the absolute contrast sensitivity reported in the original study; the left y-axis is normalized by the maximum contrast sensitivity at that eccentricity and equates to the value of the filter at the corresponding spatial frequency.

The absolute fit of achromatic contrast sensitivity data and the normalized filter curve produced by these fits are displayed in Figure 2 which demonstrates that log-Gaussian functions sufficiently described the contrast sensitivity from human data and thus were used to weight the final sum of contrast energy within each image region by the sensitivity profile of the retina in that region. The division of image regions into eccentricity bands is depicted in Figure 1b.

The weighted contrast energy (CE_w) in each image zone was then calculated using Equation 4 by taking the sum of the squared absolute values of the amplitude spectra (|X[k]|) after filtering with the appropriate log-Gaussian filter, L_gaus(f) (see Equation 2), divided by the number of sample points (N, number of pixels) in each transform:

$$\begin{array}{c}\hfill C{E}_w=\frac{1}{N}\sum _{n=0}^{N-1}{\left|X\left[k\right]L_{gaus}\left(f\right)\right|}^2\end{array}$$ (4)

Definition of eFD Regions

We have previously found that, when one eye of a young guinea pig wears a plastic ocular diffuser that attenuates the typical contrast and spatial frequency range exposure of the eye (i.e., FD), it induces the development of maximum myopia (−7 D/wk) in this species (i.e., FDM).^33,34 Similar FDM also occurs in many other species in response to diffuser wear^29–32 and in humans when vision is blocked by ptosis.⁴⁹

To determine if images taken in different visual environments (natural, mixed-urban, urban, and indoor) were potentially able to mimic FD, their local contrast (CE_w) was compared to the CE_w in a control set of images filtered through the same diffuser that causes maximum myopia in guinea pigs—Perspex, molded poly(methyl methacrylate), with a 14.5-mm diameter and 0.8-mm thickness. Any image zone in unfiltered images that had less than the maximum CE_w detected through the plastic diffuser was classified as eFD. As seen in Figure 3, these diffusers attenuate features in a scene entirely and permit 8× less contrast (CE_w) than light-perception only Bangerter foils, which induce −4.7 D of FD myopia in primates (Fig. 3)^36,50 and −4.4D of myopia in guinea pigs,³⁴ making our selected cut-off a conservative measure of potential eFD. The percentage of eFD was determined across the whole image and at each retinal eccentricity.

Figure 3.

Example mixed-urban scene (left) imaged through a light-perceptible Bangerter foil (middle) and through a Perspex ocular diffuser (right) used as the threshold of myopia in the present study. For each condition, the maximum CE_w detected is included below the image.

Assessing the Effect of Myopia-Inducing Bangerter Foils on Regional eFD

Bangerter foils result in steeper spectral amplitude–frequency gradients³⁷ and can induce myopia in animals^34–36 depending on the extent of contrast attenuation. Thus, we sought to investigate the effect of Bangerter foils on the percentage of eFD coverage observed across the visual field in different environments. The amplitude spectra of each image (n = 590) were multiplied by the modulation transfer functions (MTFs) of 0.8, 0.6, and 0.4 Bangerter foils,⁵¹ and the percentage of eFD was recalculated.

Quantifying the Complexity of Contrast Energies Across the Visual Field

If featureless manmade structures produce large regions of similarly low contrast, then the visual field within these areas would be less likely to experience a change in contrast with eye movement, sustaining the period of deprivation. To quantify this, the entropy (randomness) of the contrast energy signals received across the visual field was computed for each 8-bit contrast energy heatmap (Fig. 1g) and compared between image environments. Traditionally, the Shannon entropy of a grayscale image can be calculated using the incidence of each possible grayscale pixel. An image where each pixel value is equally probable would display maximum complexity (randomness). However, using this method to measure the entropy of 2D images ignores the spatial structure of the image and the additional simplicity produced by neighboring pixel correlations. The latter distinction is important, as the contrast experienced in neighboring image zones is most relevant when determining if contrast will change with eye movement. This limitation was overcome by applying Larkin's novel delentropy measure that uses the deldensity (p_i,j) of an image: a 2D gradient probability density function that captures the underlying spatial image structure and pixel co-occurrence.⁵² Deldensity is calculated using Equation 5:

$$\begin{array}{c}\hfill p_{i,j}=\frac{1}{4WH}\sum _{w=0}^{W-1}\sum _{h=0}^{H-1}{\delta }_{i,f_x\left(w,h\right)}{\delta }_{j,f_y\left(w,h\right)},\sum _{i=1}^I\sum _{j=1}^J{p}_{i,j}=1\end{array}$$ (5)

where f_x and f_y represent the first-order derivatives of an image in the x and y directions, W and H are the width and height of the original image, and δ is the Kronecker delta used to signify the binning operation used to determine the number of pixels in f_x and f_y matching the gradient values i and j; the sum of probabilities for the possible combinations of i and j (p_i,j) equates to 1. The possible gradient values for f_x and f_y are −255 ≥ i ≥ 255 and −255 ≥ j ≥ 255, respectively, for an 8-bit grayscale image; thus, the 2D histogram has dimensions of 511 × 511. The delentropy (H, in bits) was then calculated from the deldensity using Equation 6⁵²:

$$\begin{array}{c}\hfill H_{\left(f_x,f_y\right)}=-\frac{1}{2}\sum _{j=1}^J\sum _{i=1}^I{p}_{i,j}{log}_2{p}_{i,j}\end{array}$$ (6)

That is, by taking the sum of the products of p_i,j and their base 2 logarithms, multiplied by −1/2.⁵² Delentropy was calculated using the Sentropy package included in the ‘Simple Image Processing Pipeline’.⁵³

Statistics

The degree of eFD detected across the whole of the visual field and the complexity contrast energy heat maps were compared between image environments using one-way analyses of variance (ANOVAs) with post hoc Student's t-tests (with Bonferroni adjustment for multiple comparisons). To compare the percentage of eFD between different image environments at each retinal eccentricity, a separate repeated-measures ANOVA on ranks was performed for each environment, due to the non-normality of the data for different eccentricities. Tukey's post hoc tests were performed to adjust for multiple comparisons. To compare the effect of Bangerter foils on the percentage of eFD coverage across the visual field for each image group, a separate repeated-measures ANOVA on ranks, with Tukey's post hoc test, was performed for each environment.

Results

Manmade Environments Produce Widespread Visual eFD

The total portion of the sampled visual field that would experience eFD differed significantly among the imaged environments, F (3, 586) = 203.3, P < 0.001 (Fig. 4). Indoor environments had the highest percentage of eFD coverage, encompassing on average 29.5% ± 1.4% of the sampled visual field (Fig. 4), significantly more than that observed in outdoor urban environments (mean eFD = 15.2% ± 0.9%; P < 0.001), outdoor mixed-urban environments containing a both manmade and natural features (mean eFD = 6.7% ± 0.5%; P < 0.001), and natural environments (mean eFD = 2.8% ± 0.3%; P < 0.001). Urban environments containing only manmade structures displayed a higher percentage of eFD than mixed-urban environments (15.2% ± 0.9% vs. 6.7% ± 0.5%; P < 0.001) (Fig. 4), and both urban and mixed-urban environments displayed higher eFD than the natural environments (P < 0.001 for both) (Fig. 4). This hierarchy was consistent for all retinal eccentricities sampled, with indoor environments consistently displaying the greatest eFD, followed by urban, mixed-urban, and natural environments (Fig. 5).

Figure 4.

Manmade environments cause widespread eFD across the visual field. The percentages of eFD experienced across the visual field for each sampled image across all eccentricities from natural (blue), mixed-urban (orange), urban (yellow), and indoor (purple) environments are shown. Significance bars represent the results of post hoc Student's t-tests adjusted for multiple comparisons following significant difference with one-way ANOVA. **P < 0.001.

Figure 5.

High levels of eFD are present at all sampled eccentricities in manmade environments. The percentages of the retina at different retinal eccentricities (in degrees of visual angle) that would experience eFD in images taken in natural (blue), mixed-urban (orange), urban (yellow), and indoor (purple) environments are shown. Significance bars represent the results of post hoc pairwise tests adjusted for multiple comparisons (Tukey) following significant differences with repeated-measures ANOVA on ranks.

Diffuse manmade surfaces such as blank walls, ceilings, and roads typically contained little contrast energy within the spatial frequency range resolvable by the retina (Figs.  6c, 6e, 6g), contributing to the high number of areas classified as eFD in manmade environments (Figs.  6d, 6f, 6h, red eFD highlights). In outdoor scenes, clear portions of sky similarly displayed little contrast energy (Figs.  6a–f), contributing to the increasing vertical gradient in eFD incidence displayed in eFD heat maps from natural, mixed-urban, and urban environments (Figs.  7a, 7c). In natural and mixed-urban settings, this increase in the percentage of eFD with vertical eccentricity accounted entirely for the general increase in eFD with retinal eccentricity observed: mean ΔeFD = 3.8%, χ²(5, n = 151) = 344.5, P < 0.001 and mean ΔeFD = 8.4%, χ²(5, n = 171) = 531.0, P < 0.001, respectively (Figs.  7c, 7d).

Figure 6.

Natural and manmade elements contributing to regional eFD in different environments. (a, c, e, g) Heat maps display the CE_w values in each image zone—for example, images from natural (a), mixed urban (c), urban (e) and indoor (g) environments. (b, d, f, h) The same example images with areas classified as eFD highlighted in red.

Figure 7.

The relative distribution of eFD across the sampled visual field in different image environments. (a) Heat maps display the probability of detecting eFD in each image zone, across all images from natural, mixed-urban, urban, and indoor environments. (b) Mean percentage (markers and dashed lines) and standard error (shaded area) of the area of the retina at different retinal eccentricities (in degree visual angle averaged across all orientations) experiencing eFD for natural (blue), mixed-urban (orange), urban (yellow), and indoor (purple) environments. (c) Mean percentage of eFD coverage with vertical eccentricity (in degrees visual angle, with negative values signifying the inferior visual field and positive values signifying the superior visual field). (d) Mean percentage of eFD coverage with horizontal eccentricity (in degrees visual angle, with negative values signifying the left visual field and positive values signifying the right visual field).

Greater increases in eFD occurred with increasing eccentricity (mean of all orientations) in the indoor environments, mean ΔeFD = 16.8%, χ²(5, n = 137) = 222.7, P < 0.001 (Fig. 7b) and urban environment, mean ΔeFD = 10.2%, χ²(5, n = 131) = 183.9, P < 0.001 (Fig. 7b). This was particularly noticeable in the superior visual field (Fig. 7c). Interestingly, both indoor and urban environments also showed an increase in eFD with horizontal eccentricity, in both left and right directions (Fig. 7d). For indoor settings, examples 1 to 3 in Figure 8 account for this finding. Indoor settings restricted peripheral viewing distance, increasing the proximity of low spatial contrast containing manmade structures to the observer (Figs.  8a, 8b, 8d, 8e, 8g, 8h), typically resulting in high levels of eFD in the visual periphery where spatial resolution is lowest (Figs.  8c, 8f, 8i). Similar situations occurred in urban environments, where larger manmade structures such as buildings often occupied large proportions of the retinal periphery, similarly resulting in eFD. Example 4 demonstrates how this phenomenon did not necessarily result in eFD, if the objects positioned close to the observer contained sufficient spatial detail, such as shelved food items at a grocery store (Figs.  8k, 8l).

Figure 8.

The proximity of featureless elements to the observer in indoor environments contributes to high levels of peripheral eFD. (a–l) Four example images taken indoors (a, d, g, j), depicting the level and location of weighted contrast energy (CE_w) in each image zones (b, e, h, k) and the zones that were classified as eFD (red highlights) (c, f, i, l).

In general, the presence of featureless manmade structures resulted in large portions of the visual field experiencing similar contrast energies (Figs.  8b, 8e, 8h), meaning that eye movements would be less likely to result in a change in contrast and regional eFD was more likely to be maintained. This was assessed by applying Larkin's measure of 2D entropy (delentropy) to contrast energy heat maps from all images. For example, the delentropy of the contrast energy heat map of example 1 is 2.5 bits (Fig. 8b), 5.6× less complex than example 4 at 5.0 bits (Fig. 8h).

When assessed across all images, the delentropy of contrast energy heat maps significantly differed between environments, F(3, 586) = 161.4, P < 0.001 (Fig. 9). Unsurprisingly, contrast energy heat maps from indoor environments were the least complex with an average delentropy of 4.0 ± 0.1 bits, significantly less than those from the urban (4.5 ± 0.0 bits, P < 0.001), mixed-urban (4.8 ± 0.0 bits, P < 0.001), and natural (5.0 ± 0.0 bits, P < 0.001) environments. Contrast energy heat maps from the urban environment were also significantly less complex than those from the mixed-urban environments (4.5 ± 0.0 vs. 4.8 ± 0.0 bits; P < 0.001) (Fig. 9), and both were less complex than natural environments (P < 0.001 and P < 0.05, respectively) (Fig. 9).

Figure 9.

Contrast signals perceived across the visual field are less complex in manmade environments than in nature. The delentropy (complexity) of contrast energy heat maps from images taken in natural (blue), mixed-urban (orange), urban (yellow), and indoor (purple) environments is shown. Statistical comparisons were performed between groups using one-way ANOVA with post hoc Holm–Šídák tests, **P < 0.001.

Tasks That Restrict Eye Movements Maintain Regional eFD in Manmade Environments

In cases where eye movements are confined, such as while scanning text on a page or when using a mobile phone or computer, the probability of temporally maintaining eFD in the visual periphery would be even greater. The reading and phone examples are illustrated in Figure 10. In the first example, the low contrast energy of the desk and white paper results in a high proportion of eFD in the visual periphery (Fig. 10a). While reading, the gaze of the fovea follows the path of the red arrows (Fig. 10b) and receives contrast far above the eFD threshold (Fig. 10c), whereas the highlighted portion of the retinal periphery following the green arrow path continuously receives eFD despite movement (Fig. 10c). A similar result is observed in the second example (Fig. 10d); when the fovea is fixed on the high-contrast image of the phone screen, portions of the retinal periphery experience extended deprivation due to the featureless surround. In contrast, performing the same task in a natural environment with a detailed surround (Figs.  10g, 10h) results in the peripheral retina experiencing contrast far above the eFD threshold (Fig. 10i), even when eye movements are confined.

Figure 10.

Performing tasks that restrict eye movement indoors can result in extended periods of eFD in the periphery visual field. (a, d, g) The extent of eFD experienced while reading a page of text on a desk (a), using a phone in a typical suburban setting (d), and using a phone in a natural setting (g). (b, e, h) CE_w values for the scenes in a, d, and g, highlighting the path of the fovea while reading (red arrows) and the corresponding path of a region of the retinal periphery (green arrows). (c, f, i) The CE_w experienced by the fovea (red) and retinal periphery (green) for each example, highlighting the cut-off for classification as eFD (dashed line).

Bangerter Foils Increase the Extent of Regional eFD, Particularly in Indoor Environments

Finally, we used our novel image analysis pipeline to investigate the effect of specific contrast-reducing lens designs on the perception of eFD in different environments. Example images from natural and indoor environments filtered with the MTFs of 0.8, 0.6, and 0.4 Bangerter foils (Fig. 11a) are provided in Supplementary Figure S2. Bangerter foils caused a modest increase in the amount of eFD detected which was dependent on the degree of MTF attenuation produced by the filter (Fig. 11b). Specifically, Bangerter foils significantly increased eFD in images from natural environments, χ²(5, n = 151) = 343.5, P < 0.001 (Fig. 11b); mixed-urban environments, χ²(5, n = 171) = 508.8, P < 0.001; urban environments, χ²(5, n = 137) = 411.0, P < 0.001; and indoor environments, χ²(5, n = 131) = 393.0, P < 0.001. The largest increase in eFD was seen in indoor environments filtered with the MTFs of the strongest (0.4) Bangerter foil (+6.7% vs. unfiltered control, P < 0.001) (Fig. 11b). Interestingly, the 0.4 Bangerter foils increased eFD by only 1.2% in natural environments, equating to eFD coverage far less than even unfiltered indoor environments (4.0% ± 0.4% vs. 29.5% ± 1.4%) (Fig. 11b). Furthermore, although Bangerter foils generally reduced the complexity of contrast energy signals in all environments (Table), natural images filtered with the 0.4 Bangerter foils were still more complex than unfiltered indoor environments (4.8 ± 0.0 bits vs. 4.0 ± 0.1 bits) (Table), signifying that these filters did not completely reproduce the distorted distribution of contrast energy seen in indoor environments.

Figure 11.

Bangerter foils increase the percentage of eFD in all environments. (a) The 2D MTFs for 0.8, 0.6, and 0.4 Bangerter foils based on the data previously published by Pérez et al.⁵¹ (b) The percentage of eFD experienced across the visual field in images after multiplying the magnitude spectra of each image zone by the MTF of a 0.8, 0.6, or 0.4 Bangerter foil or none (control). Bracketed bars signify significant differences due to Bangerter foils in that group. Individual bars signify the results of Tukey’s post hoc test–adjusted multiple comparisons. *P < 0.05, **P < 0.001.

Table.

The Effect of Bangerter Foils on the Complexity of Contrast Energy Signals

	Delentropy (Bits)
	Natural	Mixed-Urban	Urban	Indoor
Control	4.95 ± 0.02	4.83 ± 0.02	4.54 ± 0.04	3.95 ± 0.06
0.8 Bangerter foil	4.85 ± 0.02	4.74 ± 0.02	4.42 ± 0.04	3.80 ± 0.06
0.6 Bangerter foil	4.82 ± 0.02	4.70 ± 0.02	4.38 ± 0.04	3.75 ± 0.06
0.4 Bangerter foil	4.79 ± 0.02	4.68 ± 0.02	4.36 ± 0.04	3.71 ± 0.06

Discussion

In the present study, we observed that the same visual signals that cause myopia in animal models are experienced across large areas of our visual field in manmade environments. Specifically, we found that feature-poor surfaces typical of manmade structures were often sufficiently devoid of contrast energy as to be classified as eFD. It was observed that, on average, 29.5% of the central 60° × 45° of our visual field would experience eFD in indoor environments compared to only 2.8% in natural environments. Urbanized outdoor spaces similarly contained a high degree of eFD, at 15.2%; however, this was more than halved (6.7%) in mixed-urban areas where greenery was integrated with manmade structures. The amount of eFD detected increased with visual eccentricity, peaking at an average of 35.3% at 27° for indoor environments; however, some images from indoor environments contained as much as 92.3% of eFD at 27°. The complexity (delentropy) of the composite of contrast energy signals received across the visual field was significantly reduced in indoor environments relative to all other settings, reflecting the increased spatial uniformity of contrast signals across the retina. This result suggests that eye movements are less likely to result in a change in regional contrast in indoor environments; thus, regional eFD is more likely to be sustained in these settings. Furthermore, the peripheral eFD and diminished contrast complexity of manmade environments mean that gaze-limiting activities such as reading or using a mobile phone are even more likely to result in extended periods of peripheral eFD. Finally, filtering image sets with the MTFs of Bangerter foils produced modest increases to the extent of eFD detected and reduced the complexity of contrast energy signals across the visual field, an effect that increased with filter contrast attenuation. Importantly, natural scenes filtered with 0.4 Bangerter foils still displayed only 4.0% eFD coverage, highlighting that these filters do not completely reproduce the distorted distribution of contrast energy seen in indoor environments. This was supported by the greater delentropy of the contrast energy heatmaps from natural images filtered with 0.4 Bangerter foils versus unfiltered indoor images (4.8 ± 0.0 bits vs. 4.0 ± 0.1 bits). The results reported here summarize features that differentiate visual signals in manmade and natural environments and suggest how our interaction with these environments has potential implications for refractive eye development.

On average, Singaporean children spend between 26 and 100 minutes a day outdoors.^6,54,55 If these children sleep for 8 hours each night, then they will spend between 14.3 and 15.6 hours a day awake indoors. During this period an average of 29.5% of their visual field would effectively experience eFD, with greater degrees of eFD in the visual periphery. Animal studies have shown that continuous regional FD results in axial elongation and myopia only in the associated portion of the visual field.^41–43 It is therefore possible that portions of the human retina experiencing eFD would elicit a similar response given enough time, suggesting that these pockets of eFD may contribute to myopia formation over many years of repeated exposure. However, the minimum stimulus size required to produce regional axial elongation and myopia is unknown, and these periods of eFD would eventually be interrupted by saccadic or voluntary eye movements.

Interestingly, the complexity of the contrast energy signals received across the whole of the retina was significantly reduced in indoor environments compared to outdoor environments. This reduction in complexity indicates that eye movements are less likely to result in different regional stimulation; thus, transient retinal activity in regions looking at featureless structures would diminish over time if eFD is maintained following movement. Assuming that all possible eye movements are equally likely, the likelihood that a portion of the retinal periphery (27°) receiving eFD would still receive eFD after eye movement is approximately 30.0% for indoor environments (i.e., the mean eFD coverage of indoor images processed using only the 27° filter). However, in cases where eye movements are confined, such as while scanning text on a page or when using a mobile phone or computer, this probability is greater. In Figure 10, we show examples of situations where the extensive peripheral eFD and diminished contrast complexity of manmade environments, mean gaze-limiting activities such as reading or using a mobile phone can result in extended periods of peripheral eFD, even though the eye is moving. These examples highlight how a combination of environmental composition and behavior can result in extended periods of regional eFD. When coupled with the reduction in mean light intensity and reduced middle and high spatial frequency content, it is perhaps unsurprising that time spent indoors is such a significant risk factor for myopia development.

The notion that the limited resolution of the peripheral retina may result in visual deprivation in certain conditions is not new. Wallman et al.⁴³ previously suggested that activities such as reading may effectively reproduce the effects of visual FD in the retinal periphery due to the large receptive fields of peripheral neurons. In the present study, the decreased filter bandwidth with retinal eccentricity did contribute to higher portions of eFD in the retinal periphery; however, it was not solely the cause of peripheral eFD. Reanalyzing images from indoor environments using the foveal filter for all areas results in a reduction of 5.3% in the degree of eFD detected at 27° (30.0% vs. 35.4%); however, the precent eFD still steadily increases by 11.7% between 5° and 27° eccentricity (18.3%–30.0%). As illustrated in Figure 8, this is likely due to the restricted peripheral viewing distance when indoors, resulting in objects containing minimal spatial detail being positioned close to the observer. Thus, eFD detected in the present study arises both due to increasing resolution constraints of the visual periphery and due to the spatial composition of manmade environments.

How, then, can interior spaces be designed or modified to emulate nature? The lobby interior of the Art Deco–inspired office building Parkview Square in Singapore is a prime example of how the selection of building materials (e.g., textured stone, patterned tiles) and intricate interior decoration can reduce the percentage of eFD to levels typical of nature (eFD = 1.3%) (Supplementary Fig. S3). However, such elaborate and expensive measures need not be taken. Simply interrupting blank spaces with pictures, art, or plants or using vinyl stickers or wallpaper to decorate a room would reduce eFD coverage. In China, some efforts have been made to increase middle and high spatial frequency contrast in primary school classrooms using detailed wallpaper,^56,57 and such modifications would appreciably reduce regional eFD when facing the front of the classroom. Encouragingly, these modifications did help to maintain the hyperopic reserve in non-myopic children compared with controls.⁵⁷ However, one could further expand upon such modifications by considering the spatial contrast present in a child's visual periphery when focusing downward on their schoolwork, whether on paper or on a digital device, as this would be more dependent on the spatial detail present on the surround such as the desk or floor.

If regional eFD imposed by indoor environments resulted in myopia formation, one would expect to first observe myopia in the visual periphery where eFD is most prevalent. Interestingly, a recent study observed that the presence of myopic refractive error in the superior retina (20° visual angle from optic nerve) in emmetropic children was highly predictive of future central myopia development.⁵⁸ These children, too, had signs of relative myopia in their inferior retina, albeit at the 16° limit of measurement. It is possible that such early refractive changes could arise from prolonged regional exposure to peripheral eFD, eventually leading to central elongation and myopia. Experiments from guinea pigs⁴⁷ and monkeys⁵⁹ have demonstrated how myopiagenic stimuli present only in the retinal peripheral can result in central axial elongation and myopia. Furthermore, if a peripheral refractive error were strong enough, it could further attenuate contrast sensitivity to higher spatial frequencies, which could increase the degree of eFD observed (like Bangerter foils), forming a positive feedback loop.

In animal models, continuous exposure to hyperopic defocus or FD can induce myopia, but even short daily interruptions can restore normal development.^50,60–62 Across species, approximately 30 min/d of lens or diffuser removal will reduce experimental myopia by 50%; however, this effect does not scale linearly, requiring 4 to 8 hours to fully attenuate the myopic stimuli.^50,60–62 Furthermore, we have previously reported that interruption without appropriate visual stimulation fails to counteract such myopic stimuli, meaning their effects can last for days.⁶² Given the phylogenetic conserved nature of this mechanism, it is likely that eFD signals are also relatively long lasting in human retina. To overcome the eFD exposures reported here, what might be most important is the time spent in alternative appropriate activities that can counteract the potential long-lasting effects of eFD, such as spending time in nature. This may not be achieved in lifestyles primarily alternating between indoor and urban settings or if behavior is dominated by extensive gaze-limiting near activities such as phone, screen, or text viewing, which consequently predisposes the peripheral retina to experience blur and contrast reduction. The exact features of counteracting visual stimuli are yet to be fully understood but would certainly include avoiding eFD.

Applying the MTF of Bangerter foils (0.8–0.4) to the amplitude spectra of our images produced a modest increase in the degree of eFD detectable in natural, mixed-urban, urban, and indoor environments. Notably, applying the 0.4 Bangerter foil MTF to images from natural scenes did not reproduce the eFD incidence of unfiltered indoor environments (4.0% for natural plus 0.4 Bangerter foil vs. 29.5% indoors control), despite previous reports that imaging natural scenes through a 0.4 Bangerter foil effectively reproduces the mean amplitude spectra of indoor environments.³⁷ This signifies that, although the mean contrast energy in each frequency band for indoor environments is reproduced by filtering natural scenes with 0.4 Bangerter foils, the distribution of such energy in the image is not. This is supported by the greater complexity of contrast energy maps from filtered natural scenes compared with those of unfiltered indoor scenes (mean difference of +0.85 bits). In general, filtering images using the MTF of 0.4 Bangerter foils produced a similar reduction in complexity across all images of ∼0.2 bits due to the compression of the signal bandwidth in all environments.

It stands to reason that the use of contrast-reducing contact lenses to treat myopia (e.g., SightGlass Vision Diffusion Optics Technology [DOT] lenses) would mimic the effects of Bangerter foils and increase the proportion of The visual field that would experience eFD. However, these lenses use only intermittently spaced dots to diffuse light. The proposed theory that these lenses could treat myopia by reducing contrast is not supported by animal studies using Bangerter foils^34,35 or environmental manipulations of contrast.³⁹ However, perhaps by diffusing high spatial frequencies located within dot regions, the total contrast energy in the perceptible frequency range in surrounding regions is increased as the resulting low spatial frequencies spill over from neighboring dot zones. This may result in a reduction in retinal eFD coverage in environments containing diffuse structures intermixed with high spatial frequency and high contrast edges, such as indoor environments. By contrast, diffuse areas already classified as eFD would be unaffected by these dots if no edges were nearby. If this were the source of treatment efficacy, then it would be dependent on dots being spaced far enough apart to scatter high-frequency edges and reduce eFD coverage in neighboring zones without inadvertently increasing total eFD coverage due to signal attenuation, as was seen here with Bangerter foils. This balance may explain why DOT lenses with scarcely spaced dots show some treatment efficacy, whereas denser arrangements have resulted in no inhibition of axial growth or myopia progression after 3 years of treatment compared with untreated controls.⁶³ If this interpretation is true, it would mean that these therapies only work because children are spending a disproportionate amount of time indoors, which raises a more general issue with the use of myopia-preventing contact lenses in natural environments, which would seemingly degrade visual signals that would otherwise normally inhibit myopia.

It should be noted that children who wore DOT lenses while reading experienced the greatest treatment effect.⁶³ Text on a page poses a unique contrast environment, characterized by a dense and relatively uniform arrangement of high-frequency black-on-white (off-center) contrast, interrupting a diffuse surface, resulting in very little contrast entropy (Fig. 11b). Perhaps attenuating such a dense arrangement of black-on-white contrast limits off-center contrast signals, underlying the treatment benefit as suggested. However, like indoor environments, diffusing the high spatial frequency contrast of text or the page edge in the visual mid-periphery may result in a net increase in contrast energy in the perceptible frequency range at that eccentricity, which is more sensitive to lower frequency contrast, or otherwise increase the entropy of contrast signals across the visual field, which may reduce regional eFD in such an environment.

Limitations

By comparing the total contrast energy within the spatial frequency range of the retina at that eccentricity to that observed through a plastic diffuser, it was simply determined if there was more or less contrast energy in that environment compared to what would be imposed on the retina by our Perspex ocular diffuser, known to induce rapid myopia development in our animal model. The advantage of this approach is that it does not assume culpability of any specific spatial frequency band; however, it produces a binary outcome: The zone is or is not eFD. This means that our assessment of contrast energy does not address whether the retina would practically be able to distinguish the difference between an image zone with contrast energy n units above the maximum contrast energy seen through our diffuser, merely that it cannot be classified as eFD based on our conservative cutoff. As such, image regions with contrast energies minimally above our threshold, which may practically be equivalent to eFD, would be excluded from our analysis, making these results conservative.

It should be noted that the filters for contrast sensitivity were normalized to represent the relative sensitivity of the retina in that region, rather than the absolute contrast sensitivity. This was intentional, as absolute contrast sensitivity is dependent on luminance, which would differ greatly between indoor and outdoor environments. The contrast sensitivity functions (CSFs) reported by Chwesiuk et al.⁴⁸ were measured at modest luminous flux, approximately reflecting the capacity of the retina to resolve contrast detail in indoor environments.⁵² Applying the same normalized CSFs to images captured outdoors therefore underrepresents the contrast sensitivity range of the retina in these settings and makes the comparison between indoor and outdoor environments more conservative. This was deemed preferable to calculating the approximate luminous flux of each image and trying to relate that to CSFs reported from multiple sources recorded using various methods at differing light intensities.

Finally, this study investigated only achromatic contrast, as the retina is more sensitive to achromatic contrast than chromatic contrast at all but the lowest spatial frequencies,⁶⁴ and chromatic contrast resolution declines more steeply with visual eccentricity than achromatic contrast.⁶⁵ As the present study was only concerned with detecting regions that sufficiently lacked contrast energy to be considered eFD, it was preferable to use only the condition that was most sensitive—achromatic contrast. However, this does produce some limitations in some extreme edge cases. For example, the grayscale image conversion used could result in the loss of chromatic contrast resulting from antiphase red–green or blue–yellow stimuli if red, green, or blue pixel intensities were equal. However, considering that FD myopia can still develop^66,67 and recover in animals reared in monochromatic light, it is likely that the primary visual cue for FD is a lack of achromatic contrast, even if longitudinal chromatic aberration does seemingly cue the sign of defocus and affects myopia development.⁶⁸

Conclusions

The present study highlights important considerations for how we design and interact with manmade structures. Diffuse manmade elements of indoor, urban, and mixed-urban environments lacked contrast, often constituting regional eFD, particularly in the visual periphery. When such environments are coupled with behaviors that restrict eye movement, such as reading or using a phone, regional eFD may be sustained due to the diminished complexity of the contrast energy profiles across the visual field. It then stands to reason that the simple prophylactic approach to myopia management of increasing outdoor time (particularly in nature) is further supported by the present study, as natural environments displayed minimal regional eFD and highly complex contrast energy profiles. However, considerations of how indoor and outdoor spaces are designed must also be made to better recreate both the spatial frequency composition and distribution seen in nature. The grocery store example highlights how indoor spaces can be just as complex as natural scenes with reduced levels of eFD. Recreating this effect in residential or school environments could be as simple as hanging more pictures, choosing more textured building materials, or using patterned wallpaper.