Quantification of Fundus Tessellation Reflects Early Myopic Maculopathy in a Large-Scale Population of Children and Adolescents

Abstract

Purpose: This study investigated the distribution of fundus tessellation density (FTD) in a Chinese pediatric population and its potential in reflecting early myopic maculopathy (tessellated fundus).

Methods: Participants were enrolled from kindergartens, primary schools, and middle schools, with cluster sampling in Shanghai, China. A series of ophthalmic examinations was conducted. Based on fundus photograph, FTD was quantitatively assessed using an artificial intelligence algorithm, and tessellated fundus was diagnosed by well-trained ophthalmologists.

Results: A total of 14,234 participants aged four to 18 years were included, with 7421 boys (52.1%). Tessellated fundus was observed in 2200 (15.5%) participants. The median of FTD was 0.86% (range 0.0–42.1%). FTD increased with age and axial length. In the logistics regression, larger FTD was independently associated with tessellated fundus (P < 0.001). The area under curves of receiver operating characteristic curve for categorizing tessellated fundus using FTD was 0.774, and the cutoff point of FTD was 2.22%.

Conclusions: The density of fundus tessellation was consistent with the severity of myopia. FTD could help diagnose the early stage of myopic maculopathy, tessellated fundus, providing a new pattern for myopia screening and detection of early myopic fundus changes.

Translational Relevance: Quantification of fundus tessellation with artificial intelligence could help detect early myopic maculopathy

Introduction

Myopia, also known as nearsightedness, is a prevalent refractive disease worldwide.¹ Myopia is becoming an epidemic in South-East Asia, with a prevalence of 80% to 90% in high school–aged children.² Myopia can lead to maculopathy, which is highly associated with bilateral visual impairment; therefore early detection of myopia is critical for the prevention of maculopathy.³ 

Fundus tessellation, the visibility of choroidal vessels observed in fundus photography, is an early myopic fundus change.⁴ Fundus tessellation warrants further investigation and change in management because it has been shown to be associated with myopic maculopathy.⁵ The current literature demonstrates that the prevalence of fundus tessellation in highly myopic populations was 94.3% (4–19 years).⁶ Apart from high myopes, the author group's recent study reported that 52.4% of a low myopic population aged nine to 12 years were with fundus tessellation.⁷ Relatively serious fundus tessellation has been defined as early myopic maculopathy, called tessellated fundus.⁸ Factors associated with fundus tessellation have been investigated in various populations; however, there is a paucity of data in children of all ages or refraction statuses, which can provide a complete picture of fundus tessellation in the pediatric population and help the management of early myopic fundus changes. 

Traditionally, fundus tessellation has been assessed by classified into several levels.⁶ With recent emergence of artificial intelligence (AI) in fundus tessellation interpretation, it can provide quantitative fundus tessellation density (FTD).⁹ With quantitative assessment by AI, the current study was to achieve a detailed description of fundus tessellation in a large-scale population of children and adolescents and to identify the potential of FTD in the screening of tessellated fundus. 

Methods

Participants

In the cross-sectional study, participants from kindergartens, primary schools, and middle schools in all 16 districts of Shanghai, China, were enrolled. The ones with amblyopia, organic eye diseases (strabismus, cataract, glaucoma, and fundus diseases other than myopic maculopathy), and intraocular surgery history were excluded. 

The study protocol was explained to all participants. Written informed consent forms were obtained from parents or legal guardians of all participants and children ≥12 years; oral consent assented from children younger than 12 years. All participants were treated with the tenets of the Declaration of Helsinki. Approval from the institutional review board of Shanghai General Hospital, Shanghai Jiao Tong University, was obtained. 

Measurement and Examinations

All participants underwent a series of examinations. Noncycloplegic refraction and corneal curvature measurements were conducted using the autorefractor (KR-8900; Topcon Optical Company, Tokyo, Japan). Each eye was measured three times. If the difference between refraction values of any two examinations was 0.50 D or greater, an additional measurement was repeated. An IOL MASTER (Carl Zeiss Meditec, Jena, Germany) was used to measure the axial length (AL). If the difference between the two measurements was 0.05 mm or greater, an additional measurement was repeated. Fundus photographs for the macula areas were acquired with a digital retinal camera (NW400; Topcon Optical Company), and the contrast, brightness, background pigmentation, and quality of the images were controlled. 

Density Quantification of Fundus Tessellation

Fundus photographs were analyzed with AI reported previously, which has been proved to be consistent with qualitative or semiquantitative methods.¹⁰ In brief, the images were preprocessed at first with the process of region of interest establishment, denoising, normalization, and enhancement. Next, the samples were labeled with automatic labeling and semi-automatic manual label correction to get the final labeled samples. Then the labeled samples were applied to the model training, with the deep learning semantic segmentation network ResnetFCN as the training model to obtain the leopard spot confidence map and the confidence probability of each pixel on the fundus belonging to exposed choroid. The model was re-trained with a separate youth dataset of 200 children and adolescents aged three to 17 years (male, n = 105, 52.5%) (Supplementary Table S1). With a well-trained model, the exposed choroid area in fundus photographs was extracted. Finally, FTD was calculated, defined as the average exposed choroid area per unit area of the fundus. 

Fundus tessellation from the fundus photographs centered on the fovea was extracted and FTD was calculated (Fig. 1). Moreover, FTD in the macula region (6 mm) was specially calculated, which were further divided into nine parts according to the Early Treatment Diabetic Retinopathy Study (ETDRS) grid centered on the fovea (center, inner superior, inner inferior, inner nasal, inner temporal, out superior, out inferior, out nasal, out temporal). 

Figure 1.

 

Density quantification of fundus tessellation. Fundus tessellation was extracted from the fundus photographs, and fundus tessellation density was calculated (A, B). ETDRS grid (6 × 6 mm) was centered on the fovea.

Diagnosis of Tessellated Fundus

Tessellated fundus was identified using color fundus photographs centered on the fovea and a similar method was adopted in our previous study.⁶ When assessing tessellated fundus, the contrast, brightness, background pigmentation, and quality of the images were all taken into account. The fundus photographs were read by two trained ophthalmologists (G.W. and J.J.) to determine the tessellated fundus. A senior ophthalmologist (X.X.) made the decision if any disagreement appeared. Intraobserver variability was tested by an ophthalmologist who randomly chose 100 of these photographs and read them twice at an interval of two weeks. Interobserver variability was tested by two ophthalmologists with another 100 randomly chosen photographs. The percentage of judgement agreement was 0.95 (kappa = 0.820), and any disagreement was discussed to reach an agreement. The percentage of intraobserver and interobserver variability were 0.98 (kappa = 0.958) and 0.97 (kappa = 0.939), respectively. 

Performance Evaluation of Segmentation Model

To test the performance of the segmentation model, fundus photographs of 50 randomly selected participants from the current cohort were used. As shown in Supplementary Table S2, the characteristics (age and myopia degree) of the population used for evaluation were representative of the whole current population. The fundus tessellation was recognized by the AI algorithm and the given FTD was compared with that obtained with manual labeling by professional ophthalmologists. Three general indicators of accuracy, sensitivity, and specificity have been calculated to evaluate the results of model. The accuracy was 0.9708, the sensitivity was 0.8879, and the specificity was 0.9799. 

Statistical Analysis

Statistical analyses were performed with SPSS (IBM SPSS Statistics, Inc., version 25.0, Chicago, IL, USA). Only data from the right eyes were chosen and included in the final analyses. The participants older than 18 years were included in the 18-years group. Continuous variables were described as the mean ± standard deviation or median and percentiles (P5, P25, P75, P95), whereas discrete variables were described as counts (proportions). Spherical equivalent (SE) was calculated as spherical power + 0.5*cylindrical power. FTD referred to the FTD of global fundus photographs centered on the fovea. MFTD referred to the FTD in the ETDRS grid centered on the fovea. 

Results

General Characteristics

In total, 14,234 participants aged four to 18 years (mean age 12.01 ± 4.17 years) were included in the analysis, whereas 260 were excluded because of amblyopia, organic eye diseases, and intraocular surgery history. The distribution of age is shown in Supplementary Figure S1. As shown in Table 1, the mean spherical equivalent (SE) was −1.88 ± 2.44 (−20.25∼7.38) D, the mean AL was 24.11 ± 1.44 (19.21∼30.46) mm, and the mean corneal curvature radius (CR) was 7.83 ± 0.27 (6.42∼9.63) mm (Table 1). The SE and AL in each age intervals are shown in Supplementary Figure S2. 

Table 1.

 

General Characteristics in the General Population, Boys and Girls

 

Tessellated Fundus and FTD

In the general population, tessellated fundus was found in 2200 (15.46%) subjects (Table 1). The median of FTD was 0.86% (P25, 0.22%; P75, 2.43%), the median of mFTD was 0.54% (P25, 0.05%; P75, 2.37%). Generally, the nasal FTD was the highest and the temporal FTD was the lowest among the regions (all P < 0.001) (Table 2). 

Table 2.

 

FTD in the General Population

 

FTD of Each Age Interval and AL Interval

The FTD of each age and AL interval was described (Supplementary Table S3). The median of FTD was 0.29% in those four years of age, and 1.88% in those 18 years or older. Figure 2A shows that FTD increased with age, and such tendency increased dramatically from 10 years. The median of FTD was 0.21% in those with AL < 21.5 mm and 4.08% in those with AL ≥ 26.5 mm. Figure 2B showed that FTD increased with AL, and such tendency increased with AL (Fig. 2). 

Figure 2.

 

FTD (quartiles) in age internals and AL internals. The quartiles (P25, P50, P75) of FTD in age internals (A) and AL internals (B) were shown in the scatter diagram. The trendlines of quartiles with age and AL internals were shown with Fit spline/LOWESS.

Around the macula, the fundus tessellation in the nasal region was the most severe. With age increasing or AL elongation, FTD in the nasal regions increased most rapidly, followed by the FTD in the inferior regions, although the development of FTD in the temporal regions was relatively slow (Supplementary Table S1). 

Distribution of FTD in Groups With/Without Tessellated Fundus

In the population with tessellated fundus, the FTD was larger. The median of FTD was 0.070 in population without tessellated fundus, and 4.295 in population with tessellated fundus. As shown in Figure 3, for the ones with tessellated fundus, there was more proportion of participants with higher FTD. 

Figure 3.

 

Distribution of FTD in population with/without tessellated fundus. The relative frequency of participants with different FTD (stepped by 1%) in population with/without tessellated fundus was shown and the values of each FTD internal were lined.

Correlation Analysis of Tessellated Fundus

Tessellated fundus was positively correlated with FTD, age, AL, CR, and negatively correlated with diopter of sphere, diopter of cylinder, and SE (all P < 0.001). Besides, girls tended to have higher tessellated fundus (P = 0.029). 

Logistics Regression of Tessellated Fundus

Logistics regression models of tessellated fundus was conducted (Fig. 4). In the analysis of tessellated fundus, after controlling gender and age, higher FTD (P < 0.001) and longer AL (P < 0.001) were the risk factors for tessellated fundus. 

Figure 4.

 

Regression analysis of tessellated fundus.

Receiver Operating Characteristic (ROC) Curve for Tessellated Fundus

The ROC for tessellated fundus with FTD was conducted. The area under curve was 0.77, and the cutoff point was 2.22, respectively (all P < 0.001). And ROC for tessellated fundus with FTD in age groups were also conducted, which showed that FTD was more effective for judging tessellated fundus in relatively old groups (Table 3). 

Table 3.

 

ROC for Tessellated Fundus With FTD in Age Groups

 

Discussion

The current study investigated a large-scale school population with a wide range of age and refractive error to assess the value, distribution pattern, and associated factors of fundus tessellation. FTD was associated with advanced age in school populations or myopia progression. The rapid progression of FTD with age started at age 10. The fundus tessellation in the nasal region showed quicker increment than temporal region of macula. Last, the FTD and tessellated fundus were significantly correlated, and FTD could be used to help diagnose tessellated fundus. 

With emergence of research demonstrating FTD to age and myopic refractive error, it has been regarded as one of most critical early signs of pathological myopia.⁸ The current study validated that FTD was positively correlated with age, suggesting that fundus tessellation was a result of accumulation with age and increasing of fundus tessellation may act as a marker in identifying visual performance, degree of myopia risk in each eye. The physiological elongation of eye axis contributed to the stretch of layers in the eye.^11,12 On the other hand, the heavy burden of school work would promote the progression of myopia and related fundus changes.^13,14 There was an acceleration phase starting from about 10 years for FTD, which slowed down at about 15 years. Thus the rapid development of FTD was during puberty, where physiological changes occur in various areas and a potential association between hormonal changes and FTD may occur. Therefore these findings signify the importance of early control and prevention of myopic fundus changes with regular screening. 

Another important element for the development of myopia could be attributed to the degree of fundus tessellation, which was consistent with the severity of myopia. With the degree of myopia becoming worse (AL increasing), the FTD significantly increased, and there was a rise in the growth rate of FTD. Besides, the increasing of FTD (macula) was more remarkable than that of the total FTD along with age. The results of the present study agreed with previous studies that reported association of fundus tessellation with severity of myopia.^15,16 Therefore children at risk of developing severe myopia should be screened early and regularly to evaluate the fundus changes and obtain appropriate management. 

Along with age increasing or myopia progression, there is a certain direction in the development of fundus tessellation, from the nasal region to the temporal region of macula. For the macula region, the development of fundus tessellation in the nasal region was relatively rapid, whereas that of the temporal region was slow. It seemed that the fundus tessellation spread from the nasal region to the temporal region. Thus fundus tessellation in the nasal region was more sensitive than that in the temporal region in the development of myopic fundus changes. Interestingly, the FTD progression in both the outer and inner nasal regions was faster than that in other regions, which suggested that the grading method of fundus tessellation should be refined according to the positions of tessellation because of their different sensibility in reflecting fundus tessellation. However, further validation of the current results is needed to verify the hypothesis. 

FTD could directly reflect the degree of fundus tessellation and help diagnose tessellated fundus. In the regression analysis, high FTD was a risk factor for tessellated fundus. According to the result of the ROC analysis, FTD is a potential marker to diagnose tessellated fundus. Additionally, taking FTD into consideration can give us a comprehensive description of fundus tessellation. As mentioned above, the judgment of tessellated fundus could just give a rough degree of fundus tessellation, mainly the position relative to macula. In contrast, the assessment of FTD could give more detailed and forewarning information. Furthermore, dividing the macula region into nine parts with an ETDRS grid and discussing FTD in each part could provide a more detailed description. 

There were several limitations in our study. First, this is a cross-sectional study with only inclusion Chinese school children; hence, the certain patterns of fundus changes cannot be determined, and future longitudinal studies are needed with inclusion of different ethnicities. Second, the inclusion of school age population, further studies should assess fundus tessellation in wide range of age. Lastly, the fundus photograph was obtained in noncycloplegic refraction and future studies should evaluated in both noncycloplegic and cycloplegic refraction. 

The current study revealed that there was an acceleration in the development of FTD starting from 10 years. The degree of fundus tessellation was consistent with the severity of myopia, with the nasal region exhibiting greater sensitivity compared to other regions of the macula. Additionally, there was more proportion of participants with respectively severe FTD in the ones with tessellated fundus, and school-age children with FTD ≥ 2.22% might be with tessellated fundus, which could be considered the key population in myopia management. The application of density quantification in assessment of tessellated fundus could simplify the diagnostic process and lowers the professional threshold, which facilitates large-scale screening of early myopic fundus changes. 

Acknowledgments

The authors thank 16 Shanghai district-level eye disease prevention and control branch centers, related community health service centers, and the schools involved for their support and many other individuals for their contributions. 

Supported by the National Key R&D Program of China (No. 2021YFC2702100; No. 2021YFC2702104; No. 2019YFC0840607); National Natural Science Foundation of China (No. 82171100; No. 81900911); Excellent Academic Leader Plan of Shanghai Science and Technology Innovation Action (No. 22XD1422900); Medical Enterprise Integration Innovation Achievement Transformation Project of Shanghai Shenkang Hospital Development Center (No. SHDC2022CRD015); Clinical Research Project of Shanghai Municipal Health Commission (No. 20214Y0427). 

Disclosure: W. Gong, None; J. Wang, None; J. Deng, None; J. Chen, None; Z. Zhu, None; I. Seth, None; B. Zhang, None; X. Wang, None; J. Yang, None; L. Du, None; X. Xu, None; X. He, None