Association between choriocapillaris perfusion and axial elongation in children using defocus incorporated multiple segments (DIMS) spectacle lenses

Abstract

Purpose

To investigate choroidal and ocular biological variables that influence axial length (AL) elongation in children wearing defocused incorporated multiple segments (DIMS) spectacle lenses.

Methods

This cohort study included 106 myopic children aged 7–14 years with a 1-year follow-up. Participants were divided into two groups according to the increase in AL in one year: rapid (>0.2 mm) and slow (≤0.2 mm) axial elongation groups. Cycloplegic autorefraction and AL were measured at baseline and after 6 and 12 months. The area of choriocapillaris flow voids (FVs) and choroidal thickness (ChT) at baseline were measured.

Results

Univariate linear regression analysis showed that AL elongation were significantly associated with the FVs area (standardised β = 0.198, P < 0.05) and age (standardised β = −0.201, P < 0.05). Multiple linear regression showed that the FVs area, age, and average K reading were associated with AL elongation. Multiple logistic regression analyses showed that greater degrees of myopia and larger FVs areas were risk factors for rapid axial elongation, while older age, large pupil diameter and steeper cornea were protective factors. In estimating axial elongation, the FVs area alone demonstrated an area under the curve (AUC) of 0.672 (95% CI, 0.569–0.775, P < 0.01), and that of FVs area and other ocular variables was 0.788 (95% CI, 0.697–0.878, P < 0.001).

Conclusion

Larger choriocapillaris FVs area at baseline may help to predict axial elongation in myopic eyes. The association between FVs area and axial elongation should be taken into consideration in further myopic cohort studies.

Introduction

The prevalence of myopia has significantly increased globally. In East and Southeast Asia, 80–90% of young adults are myopic, and the prevalence of high myopia is as high as 20% [1, 2]. High myopia is associated with a high risk of serious eye diseases such as myopic macular degeneration and retinal detachment, which can cause irreversible vision loss. Myopia has become one of the most direct causes of vision loss and an international public health problem, imposing a high economic burden [3, 4]. Therefore, studies have strived to find reliable and effective methods to control the progression of myopia.

Myopia progression is guided by visual signals [5–7]. When the image of the object is behind the retina, a condition called hyperopic defocus, the growth rate of the axial length (AL) will be stimulated and increased. When the object image is in front of the retina, a condition called myopic defocus (MD), the growth rate of the AL will slow down [7–9]. The choroid plays a crucial role in the response to retinal defocus, thickening in response to MD and thinning to hyperopic defocus [7]. Choroids are mainly composed of vascular structures, and choroidal thickening reflects an increased choroidal blood flow [10]. In animal models, increased choroidal blood flow can inhibit myopia progression [11], while decreased choroidal blood perfusion in guinea pigs leads to scleral hypoxia, ultimately associated with myopia development [12]. The choriocapillaris, the inner layer of the choroid that primarily supplies the outer retina [13], is also closely related to myopia progression [14, 15]. However, few studies have discussed the relationship between axial elongation and choriocapillaris in cohort studies.

Many special lenses have been designed to control the progression of myopia on the basis of the MD principle [16–18], including defocused incorporated multiple segments (DIMS) lenses, designed by Lam and her colleagues [19]. The central optical region of DIMS lenses is used to correct refractive errors with multiple segments of circular defocus (relative refractive power of +3.50 D) surrounding it. In clinical trials, DIMS lenses have demonstrated myopia progression control in children (62% decrease in AL within 2 years) without adverse effects on ocular visual function [20].

Few studies have analysed the factors affecting the myopia control effects of DIMS lenses or discussed the influence of choriocapillaris flow in DIMS lenses users. Therefore, we conducted a real-world retrospective study in downtown Beijing on myopic children using DIMS lenses to investigate the association between baseline ocular variables, including choriocapillaris flow, and 1-year myopia progression.

Materials and methods

The Ethics Committee of Peking University People’s Hospital approved this retrospective cohort study, which we conducted according to the ethical standards established in the Helsinki Declaration.

Study participants

The participants were allocated to wear DIMS spectacle lenses in the Peking University People’s Hospital Eye Center from September 2019 to September 2020.

Inclusion criteria:

1. Residency in downtown Beijing, China

2. Baseline age of enrolment: 7–14 years.

3. Cycloplegic spherical equivalent refraction (SER): –0.75 to –5.00 D.

4. Anisometropia and astigmatism: ≤1.50 D.

5. Best corrected visual acuity: 20/20 (0.00 logMAR) or higher.

6. Available complete follow-up data: corneal topography and optical coherence tomography angiography (OCTA) examination of the fundus before starting to wear DIMS lenses; AL and cycloplegic SER measured at baseline and after 6 and 12 months of daily DIMS lenses use.

Exclusion criteria:

1. Strabismus and binocular vision abnormalities, including abnormal binocular combination, interocular interaction and stereopsis.

2. Any ocular or systemic diseases and abnormalities that may affect visual function or refractive development.

3. Previous experience with other myopic control methods, such as orthokeratology, additional progressive lenses, bifocal lenses, and medication treatment, such as low concentrations of atropine.

The main follow-up variable was AL, a more repeatable and sensitive measure than cycloplegic SER and the preferred endpoint for assessing myopia progression [21]. To facilitate effective clinical prediction of the control effect of the DIMS lenses, we divided participants into two groups according to their AL changes in one year: a rapid group, with AL increases larger than 0.2 mm, and a slow group, with AL increases smaller than or equal to 0.2 mm [22, 23], corresponding to different myopia control trials.

Ocular biological variable measurement

To prevent overestimation of myopia, cycloplegic refraction was measured using a Nidek ARK-510A (Nidek, Japan) 30 minutes after the administration of three drops of 0.5% compound tropicamide eye drops (Santen Pharmaceutical Co. Ltd, Japan; 0.5% tropicamide combined with 0.5% phenylephrine) given 5 min apart. Cycloplegic SER was calculated as the sphere + 0.5 cylinder, and AL was obtained by an IOLMaster system (Carl Zeiss, Germany). Both cycloplegic SER and AL were measured at baseline and 6 and 12 months of wearing DIMS lenses. In each visit, the physician checked if the myopia and astigmatism were fully corrected. If they do not match, replace the lenses with fully corrected ones. At the same time, the position of the lenses would be checked to ensure the correct use of DIMS lenses. All measurements were repeated five times, and the average value was calculated.

We measured the corneal topography using a Sirius Scheimpflug photography-based system (CSO, Firenze, Italy) to obtain the corneal average K reading (Ave-K), central corneal thickness (CCT), anterior chamber depth (ACD), and pupil diameter (PD). PD was measured under the same light intensity (photopic model, 40 lux) each time after dark adaptation for five minutes, and the examination was performed according to the instrument’s protocol.

Choroidal biomarkers

In order to avoid the potential effect of cycloplegia on the baseline choroid parameters, choroidal parameters were scanned with natural pupils and no caffeine taken for the 24 h before choroidal imaging. Before the screening, the participants were directed to have a 10-minute distance viewing [24] with full distance sphere and astigmatism correction, in order to eliminate the effects of previous environmental [25], high accommodation, and optical factors [26–28]. All measurements of choroid were completed between 1 p.m. and 4 p.m. We measured all choroidal parameters using spectral-domain optical coherence tomography angiography (SD-OCTA) with a RTVue XR OCT with AngioVue version apparatus (Optovue, Freemont, CA, USA), which operates at a high speed of 70000 scans per second and a wavelength of 840 nm based on split-spectrum amplitude-decorrelation angiography with 5-micron axial and 15-micron transversal resolutions. The OCTA procedure included 6 × 6-mm scans centred on the fovea with 400 × 400 A-scans.

The choriocapillaris layer was sampled as a 30 μm-thick slab starting from the posterior to the retinal pigment epithelium–Bruch membrane complex segmentation. Quantifying the choriocapillaris flow directly is difficult due to the large illumination spot size and the insufficient lateral resolution for visualising the choriocapillaris clearly in the posterior pole; therefore, we used the flow void (FVs) area to identify and assess the status of the choriocapillaris [29].

To avoid the impact of axial length variation on the choroidal measurement, the Littman formula and the modified Bennett formula were used to corrected the magnification error of the image [30, 31],which should be corrected by: Dt²/Dm² = 0.002066 (AL– 1.82)².In the formula, Dm was measured OCT image diameter, Dt = 23.82 mm (true fundus diameter according to the Bennett formulae), and AL is the axial length in mm, 1.82 is the distance in mm from the corneal vertex to the second principal plane.

The choriocapillaris and superficial retina layer images were transformed into binary images and processed by ImageJ FIJI (version 1.51a, available at Fiji.sc free of charge). For the superficial retina layer (Fig. 1A), we applied the Yen threshold (Fig. 1B) to calculate retinal vascular artefacts (Fig. 1E) [32]. For the choriocapillaris layer (Fig. 1C), we calculated an automatic local threshold with the Phansalkar method (radius = 2 pixels) (Fig. 1D) to calculate the FVs area (Fig. 1E) [33, 34]. The particle area in a 3-mm centred on the fovea (Fig. 1F) was analysed. The actual FVs area (black pixels in Fig. 1F) was the choriocapillaris layer FVs area minus the retinal vascular artefact area.

Fig. 1. Calculation of the flow void area.

A The image of the superficial retina was converted to 8-bit and B subjected to Yen threshold processing. C The choriocapillaris layer was converted into an 8-bit image and D subjected to auto-local thresholding (Plansalkar’s Method, radius = 2). E The particle area in a 3-mm centred on the fovea was analysed. FThe calculation focused on an area of central 3 mm circle surrounding, and the black particle area was analysed.

We measured the Subfoveal choroidal thickness (SFChT) in the fovea using the manufacturer’s ruler in the built-in software. SFCT was defined as the vertical distance from the hyperreflective line of the Bruch’s membrane to the hyperreflective line of the inner surface of the sclera.

To affirm agreement within and between examiners, the intra-visit and inter-visit reproducibility was examined in a considerably smaller study group consisting of 20 right eyes of myopic volunteers (10.2 ± 2.03 years,10 men), who did not participate in the cohort study and were scanned 3 times with 1 min breaks between each measurement. Intraclass correlation coefficients (ICCs) and Bland–Altman plots were used to assess the agreement of SFChT within and between examiners. Due to the automatic calculation of ImageJ software, ICCs and Bland–Altman plots were only used within examiners to assess the agreement of FVs.

Statistical analysis

Only the right eye was used for statistical analysis. We used descriptive statistics to describe the baseline characteristics of the study participants. Unpaired t tests and binary logistic regression were used to analyse the association between different groups and ocular biological variables. Correlation factors were analysed using univariate analysis and multivariate analysis via linear regression. We calculated the area under the receiver operating characteristic (ROC) curve to evaluate the FVs area and multifactor predictive control on myopia control progression. SPSS software (version 26.0) was used for statistical analyses, and two-tailed P values lower than 0.05 were considered statistically significant.

Results

A total of 106 patients met the criteria for inclusion in the study. Among them, 38% (n = 39) demonstrated a good control effect, and 62% (n = 67) had rapid growth of myopia. There were no significant differences in SER or AL at baseline between the two groups. The ICCs values of SFChT was 0.933 and 0.973 (Supplementary Tables S1) within and between examiners, and Bland–Altman plots (Supplementary Fig. S1) showed good intraobserver and interobserver agreement for SFChT. For FVs area, high ICC values (Supplementary Table S2) and Bland–Altman plots (Supplementary Fig. S2) indicated strong repeatability.

The average ChT was 255.86 ± 35.29 µm (169–339 µm) and the mean FVs area was 2.25 ± 0.22 mm² (1.76–2.91 mm²). The baseline characteristics of all patients are shown in Table 1. The mean baseline FVs area of the rapid progression group was larger than that of the slow progression group (2.30 ± 0.23 mm² vs. 2.17 ± 0.18 mm²). There was no significant correlation between the FVs area and other ocular and general variables except ACD (r = 0.236, P < 0.05). There was also no significant difference in other ocular biometers between the rapid progression group and slow progression group.

Table 1.

Participants’ ocular biometric and choroidal variables at baseline.

Parameters	All participants (N = 106)	Rapid group (N = 67)	Slow group (N = 39)	P value
Age at enrolment (years)	10.21 ± 1.69	9.98 ± 1.61	10.42 ± 1.77	0.19
Gender, male (n)	51 (55)	30 (37)	21 (18)	0.37
Ave-K (D)	42.98 ± 1.64	42.84 ± 1.86	43.21 ± 1.18	0.82
CCT (µm)	554.07 ± 33.39	556.43 ± 34.73	550.12 ± 31.10	0.36
ACD (mm)	3.31 ± 0.28	3.33 ± 0.30	3.26 ± 0.23	0.21
PD (mm)	4.16 ± 0.61	4.11 ± 0.60	4.24 ± 0.64	0.32
SER-baseline (D)	–2.44 ± 1.29	–2.47 ± 1.23	–2.40 ± 1.39	0.82
AL-baseline (mm)	24.62 ± 0.77	24.56 ± 0.81	24.72 ± 0.69	0.46
ChT (µm)	255.86 ± 35.29	255.73 ± 37.10	256.07 ± 32.41	0.96
FVs area (mm2)	2.25 ± 0.22	2.30 ± 0.23	2.17 ± 0.18	<0.01

D dioptre, Ave-K corneal average K reading, CCT central corneal thickness, ACD anterior chamber depth, PD pupil diameter, SER spherical equivalent refraction, AL axial length, ChT choroidal thickness, FVs choriocapillaris flow voids.

After 1 year, patients wearing DIMS lenses (Table 2) in the slow progression group had a 1-year average increase in AL of 0.13 ± 0.06 mm, and those in the rapid progression group had a 1-year average increase in AL of 0.32 ± 0.09 mm (F = 149.18, P < 0.001), while the total SER increases were –0.38 ± 0.40 D and –0.58 ± 0.34 D, respectively (F = 7.29, P < 0.01). Children with slow myopia progression presented with a 39% slower SER progression than children with rapid myopia progression, and the AL growth rate decreased by 76% in one year.

Table 2.

Cycloplegic spherical equivalent refraction and axial length changes in participants at the 1-year follow-up.

	Rapid group (N = 67)	Slow group (N = 39)	Mean difference
Time/visit	Changes in SER (D)
6 months	–0.36 ± 0.29	–0.24 ± 0.26	–0.32 ± 0.28*
12 months	–0.58 ± 0.34	–0.38 ± 0.40	–0.51 ± 0.37**
Time/visit	Changes in AL (mm)
6 months	0.15 ± 0.09	0.06 ± 0.12	0.12 ± 0.11***
12 months	0.32 ± 0.09	0.13 ± 0.06	0.25 ± 0.12***

Statistically significant difference between two experimental groups (unpaired t tests, *P < 0.05, **P < 0.01, ***P < 0.001.).

Linear regression analysis of axial elongation and baseline factors

Univariate linear regression analysis (Table S1) showed that axial elongation was significantly associated with the FVs area (standardised β = 0.198, P < 0.05) and age (standardised β = − 0.201, P < 0.05), as shown in Fig. 2.

Fig. 2. Linear regression analysis betwee axial elongation, age and flow void.

Scatter plots showing the correlation between A axial elongation and FVs and B axial elongation and age.

In stepwise linear regression, the relationships among CCT, ACD, ChT, age, ave-K, FVs, SER, PD and with axial elongation were calculated. The result showed that age (standardised β = −0.237, P < 0.05), FVs (standardised β = 0.227, P < 0.05) and Ave-K (standardised β = −0.243, P < 0.05) were independently associated with AL.

Factors influencing myopia control in children wearing DIMS lenses in logistic regression

We analysed the association between the axial elongation effects and eye variables in patients with myopia wearing DIMS lenses by univariate and multivariate binary logistic regression analyses. In our univariate analysis, a higher FVs area (OR, 21.337; 95% CI, 2.549–178.604, P < 0.01) was a risk factor for a rapid AL change after treatment to prevent myopia progression (Table S2).

Backward selection to variables (CCT, ACD, ChT, age, ave-K, FVs, SER and PD) by logistic regression to develop a risk model for axial elongation. The regression analysis demonstrated that FVs was an independent predictor of axial elongation (OR = 75.612, 95% CI: 5.991~954.354, P < 0.001). In addition, age (OR = 0.631, 95% CI: 0.458~0.870, P < 0.01), Ave-K (OR = 0.720, 95% CI: 0.524~0.989, P < 0.05), PD (OR = 0.448, 95% CI: 0.211~0.951, P < 0.05) and SER (OR = 0.659, 95% CI: 0.439~0.990, P < 0.05) were independently associated with axial elongation in 1 year.

Analysis of ROC curves of the FVs area alone and FVs area with other ocular variables in predicting axial elongation

Figure 3 shows the ROC curve of the FVs area alone (green). The optimal cutoff value for the FVs area was 2.32 mm², and the sensitivity and specificity were 46.3% and 83.3%, respectively. Which means relatively weak ability to diagnose with rapid AL elongation. Our results show that the area under the curve (AUC) of the FVs area alone in predicting axial elongation was 0.672 (95% CI, 0.569–0.775, P < 0.01).

Fig. 3. ROC Analysis to predict axial elongation.

ROC curve of FVs area alone and total equation (including FVs, age, SER, Ave-K and PD) to predict axial elongation in children wearing DIMS lenses.

The ROC curve of the FVs area combined with ocular variables (PD, Ave-K, and SER at baseline), and age is shown in Fig. 3 (blue). The figure shows that the AUC was 0.788 (95% CI, 0.697–0.878, P < 0.001), and the sensitivity and specificity of the model in estimating the probability of myopia were 69.2% and 85.1%, respectively, which was relatively more accurate to predict rapid AL elongation compared the FVs only.

Discussion

In our study, we retrospectively analysed one-year changes in AL and SER values of patients wearing DIMS lenses in a real-world study, and we established a one-year AL change predictive model based on the baseline data. We found that the FVs area was a potential predictor of axial elongation when wearing DIMS lenses; specifically, a smaller FVs area, which reflects more sustained choriocapillaris perfusion, may imply slower myopia progression. We also found that for children wearing DIMS lenses, older age, initially less myopic eyes, larger pupil size, and steeper corneal curvature were protective factors for myopia control effects.

Myopia control effects of DIMS lenses

In our retrospective study, the average one-year AL increase with DIMS lenses was approximately 0.25 mm. The control effect is similar to that reported after administration of 0.01% atropine alone (0.30 mm/year) [35] and after use of Ortho-K alone (0.20 mm/year) [36]. In the study by Lam et al. [19, 37], DIMS lenses had a good control effect, slowing AL progression by 62% within two years with an average annual increase of 0.11 mm, which is similar to the progression of the slow group in this study. For the participants with single vision lenses in Lam’s study, the average annual increase in AL was 0.28 mm, which is similar to the progression of the rapid group in this study. However, this rate is still slower than that of myopic children without intervention [35, 38]. In addition, the education burden in Beijing is relatively large. Children spend more time focused on near vision activities than other children. Especially during the COVID-19 pandemic, the study time spent relying on video terminals increased, and the stay-at-home rules decreased the time spent on outdoor activities.

Analysis of demographic and ocular biology variables affecting axial elongation

Our study analysed factors influencing the myopia control effect of DIMS lenses. We found that older age was a protective factor for myopia control and myopia at baseline was a risk factor, findings similar to those in previous studies and epidemiological surveys [39, 40].

In addition, our study shows that a large pupil size under natural conditions is a protective factor for the effectiveness of defocus lenses in controlling myopia progression. This may be closely related to the design of the DIMS lenses. A larger natural pupil results in a larger peripheral, multifocal zone with multiple segments of relative peripheral myopia on the retina. In animal experiments, more MD during visual development resulted in better myopia control [7, 41]. A similar conclusion was found in patients who used Ortho-K [42]. Pupil size plays an important role in allowing more peripheral light rays to reach the retina.

Steep corneas were associated with less axial elongation in the current study, suggesting beneficial effects from myopic defocus. However, in Chinese children in a study from Singapore, the corneal radius was negatively correlated with peripheral spherical equivalent at temporal and nasal eccentricities of 30° [43]. A steeper cornea was related to a less myopic spherical equivalent at the temporal and nasal 30° eccentricities, which may mean more relative peripheral hyperopia. A higher amount of peripheral hyperopia or less myopic defocus in the peripheral retina would not be beneficial to a myopia control effect. Given this discrepancy and that the effect of corneal power was not significant in the current study’s univariate analysis, the effect of corneal power seems minor. We also did not measure peripheral defocus directly in our study to obtain stronger evidence; thus, further study is required.

The choriocapillaris could potentially predict myopia progression

The choroid plays a mediating role in the signalling pathway of ocular growth from the retina to the sclera, which is closely associated with the progression of myopia. In animal models of myopia, such as guinea pigs [11, 44] and chicks [45, 46], both form-deprivation myopia and lens-induced myopia studies showed that MD increased the choroid thickness. Similar results were found in humans [47], suggesting that MD may increase choroidal blood flow to decrease AL progression after a hyperopia-inducing stimulus. Hence, it is of great significance to evaluate choroidal blood flow in the progression of myopia and its efficacy for myopia prevention and control.

In previous studies, the ChT in the fovea [48], luminal choroidal area (LCA) and stromal choroidal area (SCA) [49] and the choroidal vascularity index (CVI) [50] have been used to quantify the choroidal blood flow in the medium-large layer. These indicators are limited by choroidal circadian rhythms, but choriocapillaris perfusion showed robustness during the analyses [51]. In our study, we used the choriocapillaris as the main analysis layer due to its good stability and robustness. The choriocapillaris contains the terminal vessels of the choroid, which directly supply nutrients to retinal photoreceptor cells [7]. Whether the photoreceptors are adequately nourished and supported is of great importance in the development of myopia [52].

To the best of our knowledge, we found for the first time the predictive effect of the FVs area on myopia control in myopic children wearing DIMS lenses, suggesting that a low choriocapillaris flow at baseline may contribute to rapid myopia progression. This finding is consistent with the conclusion obtained in guinea pigs [53], stating that better choriocapillaris perfusion is associated with myopia control effects, while lower choriocapillaris perfusion corresponds with more AL progression. We reached the same conclusion in this human defocus study.

Near-work activities during accommodation influence choroidal dimensions. Woodman et al. [54] found that the ChT of myopic eye decreased significantly during accommodation. Chang et al. [55] reported that only myopic eyes showed significant changes in choroid blood in accommodation. Greater accommodation-induced choroid thinning was related to the retinal OFF-Pathway overstimulation [56]. In myopic clinical studies, it is usually hard to control the different near-work hours between different participants, which may lead to the research bias. Choroidal thickness and volume were related to accommodation, which may alter the impact of near work on myopia progression. Analysing FVs in addition to choroidal dimensions from OCT may provide further insights regarding AL elongation.

In studies of high myopia in adults, there is a greater FVs area in individuals with high myopia than in those with low or moderate myopia [57]. In our study, in younger adolescents, a greater FVs area at baseline indicated more axial growth during the next year. The causal relationship between the FVs area and the progression of myopia remains to be further elucidated, and we still do not know whether a lower FVs would lead to rapid myopia progression or vice versa.

Limitations

Our study has some limitations. First, ours was a retrospective, single-centre, and single ethnic background study. All participants wore DIMS lenses, and the factors that may influence the myopia control effect for DIMS lenses should be verified in future studies concerning other treatments, i.e., single vision spectacle lenses and orthokeratology lenses. Second, the sample size of the study was relatively small, and we excluded children with high myopia. In the future, a larger cohort study is needed to determine the threshold of choriocapillaris perfusion for predicting myopia progression after wearing DIMS lenses. Finally, there are technical limitations to OCTA, including motion or shadow artefacts. For example, superficial retinal vessels may confound choriocapillaris analysis, even though we used image processing techniques to mask the area beneath the main superficial vessels.

In conclusion, we explored the use of DIMS lenses to control myopia progression in affected children. A larger PD, older age, steep corneal radius, less myopia at baseline, and smaller FVs area were protective factors against rapid axial elongation in children wearing DIMS lenses. In addition, we built a control effect prediction model for DIMS lenses. The FVs area has potential as a predictor of myopia progression for DIMS lens wearers.

Summary

What was known before

• The prevalence of myopia has significantly increased globally.

• The defocused incorporated multiple segments (DIMS) lenses have been designed to control the progression of myopia effectively.

• Predicting the effect of myopia control is of great significance.

What this study adds

• In children wearing DIMS lenses, We found that (A) larger pupil diameter and older age were protective factors, (B) but more choriocapillaris flow voids and high myopic eyes were risk factors against myopia progression.

• We build a control effect prediction model for DIMS lenses to predict myopia progression.

• Choriocapillaris flow voids have the potential13 to become a predictor of myopia progression for DIMS lens wearers.

Author contributions

XL and JH participated in the conception and design of this work, data acquisition and analysis, literature search and manuscript writing. ZP contributed to data collection and data analysis. SC and LS contributed to data collection, the acquisition of fundus photographs and picture processing of the datasets. YL and MZ contributed to the interpretation of the data and revision of the manuscript for this study. KW contributed to the design of the work, interpretation of the data, and revision of the manuscript for this study. All authors read and approved the final version of the manuscript.

Funding

This work was supported by the Capital’s Funds for Health Improvement and Research (No. 202-1G-4083), the National Natural Science Foundation of China (Grant Nos. 82171092, 81870684), the National Key R&D Program of China (Nos. 2021YFC2702100, 2020YFC2008200).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41433-023-02629-2.