Machine learning based strategy surpasses the traditional method for selecting the first trial Lens parameters for corneal refractive therapy in Chinese adolescents with myopia

doi:10.1016/j.clae.2020.05.001

Contact Lens and Anterior Eye

Volume 44, Issue 3, June 2021, 101330

https://doi.org/10.1016/j.clae.2020.05.001 Get rights and content

Abstract

Purpose

Return zone depth (RZD) and landing zone angle (LZA) are important parameters of corneal refractive therapy (CRT) lenses. A new machine learning algorithm is proposed for prescribing CRT lens parameters in Chinese adolescents with myopia.

Methods

This is a retrospective study. In total, 1037 Chinese adolescents with myopia (1037 right eyes) were enrolled. A calculation model based on corneal elevation maps was constructed to calculate RZD and LZA for the four quadrants. Furthermore, multiple linear regression and optimized machine learning models were established to predict RZD and LZA values for different combinations of age, sex, and ocular parameters. The four methods (sliding card, linear regression, calculation and optimized machine learning) were then compared to the parameters of the final ordered lens.

Results

The optimized machine learning pipeline achieved the best performance. Age, sex, horizontal visible iris diameter (HVID), spherical equivalent refraction degree (SER), eccentricity (e), keratometric (K) readings, corneal astigmatism (CA), axial length (AL), AL/corneal curvature ratio (AL/MK), and anterior chamber depth (ACD) were significant to the machine learning model. The R values for the nasal, temporal, superior and inferior LZA based on machine learning were 0.843, 0.693, 0.866 and 0.762, respectively, and those for the RZD were 0.970, 0.964, 0.975 and 0.964, respectively.

Conclusions

The feasibility and efficiency of an optimized machine learning method to predict LZA and RZD parameters has been demonstrated. The advantage of the proposed method is that it is more accurate, easier to use and faster to implement than the traditional sliding card method.

Introduction

Myopia is an ametropic eye disease with high incidence and rapid progression in children and adolescents worldwide. [1] The incidence of myopia in young adults is approximately 80–90%, and there is a high incidence of high myopia in young adults (10–20%) in east and southeast Asia [2]. In recent years, a myopia boom has been observed in China, and the onset of myopia has been diagnosed in much younger children [3]. While spending approximately three hours per day under an illuminance of at least 10,000 lx is an effective way to prevent myopia progression [4], Chinese adolescents spend little time per day on outdoor activities because of a heavy homework load; the only other strategies for myopia control include orthokeratology (ortho-k) treatment, multifocal lenses and low-concentration atropine eye drops, and the latter two methods have not yet been approved by the Chinese FDA.

Ortho-k treatment plays an important role in myopia control, [5] and many conclusive studies have shown that it can effectively slow axial length (AL) elongation in children with myopia [6,7]; quite a few of these studies were randomized controlled trials (RCTs) and systematic reviews [8,9]. Corneal refractive therapy (CRT) lenses from Paragon, which have recently been approved by the Chinese FDA, consist of three basic parameters: the base curve (BC), the returning zone depth (RZD) and the landing zone angle (LZA). When fitting CRT lenses, corneal topography and cycloplegic refraction examinations are very important for targeting these three parameters [10]. The parameters of the adjacent three zones are relatively independent based on the total lens sagittal height and are easy to adjust.

Currently, the most common method for choosing the first CRT trial lens is based on a sliding card provided by the manufacturer. [11] This method requires only two parameters to select the first trial lens—the flat K reading and the spherical reduction needed to correct myopia—and does not provide any suggestions for toric lenses [11]. However, depending on the relationship between ortho-k lens design and corneal morphology, many factors in addition to the flat K reading and spherical reduction need to be considered. Eccentricity (e) values may have a great influence on RZD and LZA values. The accuracy of the lens fit decreases as the cornea exhibits a steeper (flatter) curvature associated with a higher (lower) e value [12]. The larger the e value is, the shallower the periphery of the cornea, and the smaller the required LZA value. For the same reason, one may expect a patient with a deeper anterior chamber depth (ACD) value and a steeper cornea to need a deeper RZD and a steeper LZA. However, the sliding card is based on the corneal features of Western adolescents, which differ from those of Chinese subjects [13], and there are statistically significant differences between CRT parameters for the first trial lens suggested by the nomogram and the lenses that are ultimately prescribed [11]. In summary, the sliding card method is very simple but may not be accurate for Asians, and repeated lens trials may be needed to obtain the best parameters, which may result in a poor clinical experience for patients. Furthermore, multiple trials may increase the risk of corneal epithelial damage.

In previous studies, machine learning algorithms have been widely applied in optometry and ophthalmology for such purposes as classifying keratoconus, performing medical imaging, and predicting myopia. [14,15] A machine learning algorithm normally starts with the system calculating the image features that are believed to be important in making the prediction or diagnosis [16]. However, there is no relevant report on the application of machine learning in the field of lens selection for CRT ortho-k lenses, a process that is of great significance for the effective and accurate selection of the lens parameters. In this study, a machine learning method is applied to select CRT lenses. Additionally, this is a novel investigation of the use of a machine learning method to predict the first CRT trial lens for Chinese adolescents with myopia during which a very precise, optimized machine learning model was established to better determine the final parameters.

Section snippets

Clinical characteristics

This retrospective study analyzed data collected from 1037 eyes of 1037 Chinese adolescents with myopia who were fitted with Paragon CRT lenses for myopia management at the Ophthalmology Department and Optometry Center of Peking University People's Hospital from January 2016 to December 2018. The inclusion criteria were 1. myopic patients who were seeking ortho-k treatments with SER more than -0.38D; 2. ocular astigmatism was no more than -3.50 D; 3. best corrected visual acuity (BCVA) values

Estimated RZD and LZA values in the four quadrants

In total, 1037 LZA and RZD observations were analyzed. The means of the LZA achieved by the calculation model in the nasal, superior, temporal, and inferior quadrants were 31.87 ± 1.24° [95% confidence interval (CI) 31.79°-31.94°], 31.71 ± 1.74° [95% CI 31.60°-31.81°], 33.03 ± 1.17° [95% CI 32.96°-33.10°], and 33.04 ± 1.33° [95% CI 32.96°-33.12°], respectively (Fig. 4). There were significant differences in the LZA among the four quadrants (P < 0.0001). There were no significant differences in

Differences in corneal morphology between chinese and western patients

Ortho-k treatment has been proven to be a safe and efficient way to slow ocular axial elongation in children with myopia; it can slow the growth of the AL by approximately 30%∼80% per year. [[24], [25], [26]] The lens-fitting process is critical for prescribing appropriate lens parameters, which depend on the corneal morphology of individuals. However, corneal morphology varies across ethnic groups. Previous studies have observed that Chinese people tend to have flatter, more aspheric corneas

Conclusion

In conclusion, the optimized machine learning method shows good performance for estimating the parameters of the first trial CRT lens and outperforms the traditional sliding card method and the traditional multiple linear regression model in both accuracy and efficiency. This optimized model could be applied to achieve more personalized lens design in the future.

Funding support

This work was funded by the National Natural Science Foundation of China (Grant No. 81870684 & 81421004), the HuaXia Translational Medicine Fund for Young Scholars (Grant No. 2017-B-001), the Non-profit Central Research Institute Fund of Chinese Academy of Medical Sciences (Grant No. 2019HY320001), and the National Key Instrumentation Development Project of China (Grant No. 2013YQ030651).

Declaration of Competing Interest

The authors have no conflicts of interest to disclose and no proprietary interests in any of the materials mentioned in this article.

Acknowledgements

References (36)

I.G. Morgan et al.
The epidemics of myopia: aetiology and prevention
Prog Retin Eye Res
(2018)
M.Clara Arbelaez et al.
Use of a support vector machine for keratoconus and subclinical keratoconus detection by topographic and tomographic data
Ophthalmology
(2012)
L. Zhang et al.
A novel fitting algorithm for alignment curve radius estimation using corneal elevation data in orthokeratology lens trial
Contact Lens Anterior Eye
(2017)
S.M. Li et al.
Efficacy, safety and acceptability of orthokeratology on slowing axial elongation in myopic children by meta-analysis
Curr Eye Res
(2016)
J. Huang et al.
Efficacy comparison of 16 interventions for myopia control in children: a network meta-analysis
Ophthalmology
(2016)
Y. Zhang et al.
Corneal stiffness and its relationship with other corneal biomechanical and nonbiomechanical parameters in myopic eyes of chinese patients
Cornea
(2018)
D.S.W. Ting et al.
Deep learning in ophthalmology: the technical and clinical considerations
Prog Retin Eye Res
(2019)
A.Z. Crawford et al.
Comparison and repeatability of keratometric and corneal power measurements obtained by Orbscan II, Pentacam, and Galilei corneal tomography systems
Am J Ophthalmol
(2013)
S. Quisling et al.
Comparison of Pentacam and Orbscan IIz on posterior curvature topography measurements in keratoconus eyes
Ophthalmology
(2006)
D.O. Mutti et al.
Myopia: an epidemic of possibilities?
Optom Vis Sci
(2015)

C. Ching-Yu et al.

Refractive errors in an elderly chinese population in Taiwan: the shihpai eye study

Invest Ophthalmol Vis Sci

(2003)

E. Dolgin

The myopia boom

Nature

(2015)

C. Connie et al.

Myopia control using toric orthokeratology (TO-SEE study)

Invest Ophthalmol Vis Sci

(2013)

S. Yuan et al.

Orthokeratology to control myopia progression: a meta-analysis

PLoS One

(2015)

H. Takahiro et al.

Long-term effect of overnight orthokeratology on axial length elongation in childhood myopia: a 5-year follow-up study

Invest Ophthalmol Vis Sci

(2012)

J.J. Nichols et al.

Overnight orthokeratology - response

Optom Vis Sci

(2000)

H.A. Swarbrick

Orthokeratology review and update

Clin Exp Optom

(2010)

D.J. Gormley et al.

Corneal modeling

Cornea

(1988)

Cited by (10)

Exploring the relationship between urban residents' emotional changes and built environment before and during the COVID-19 pandemic from the perspective of resilience
2023, Cities
During the COVID-19 pandemic, the emotion of urban residents has been largely affected, resulting in different levels of stress and negative emotions. Emotion is an indispensable component of urban resilience, especially in the post COVID-19 period. The changes in residents' moods have the potential to reflect the degree of resilience of urban space as a carrier of daily urban life. This study used machine learning to analyze residents' emotion changes revealed by their posts in Sina Weibo mobile application, and analyzed urban resilience through exploring the relationship between urban built environment and emotional changes before and during the pandemic, using Shenzhen as an example. Our results illustrated the spatial pattern of emotional changes in Shenzhen, and we found that built environments could enhance urban resilience in two approaches. On one hand, catering, retail and sports facilities are supportive to emotional stability and thus contribute to the relatively steady emotional changes. On the other hand, proximity to open space is related with positive emotion, and thus contributes to higher levels of restoration. In contrast, industrial and urban village land uses are associated with lower levels of emotional resilience. This study explored to assess urban resilience from the perspective of emotional changes, which would provide insights for environmental interventions in the post-pandemic era.
A machine learning-based algorithm for estimating the original corneal curvature based on corneal topography after orthokeratology
2023, Contact Lens and Anterior Eye
To estimate the original corneal curvature after orthokeratology by applying a machine learning-based algorithm.
A total of 497 right eyes of 497 patients undergoing overnight orthokeratology for myopia for more than 1 year were enrolled in this retrospective study. All patients were fitted with lenses from Paragon CRT. Corneal topography was obtained by a Sirius corneal topography system (CSO, Italy). Original flat K (K1) and original steep K (K2) were set as the targets of calculation. The importance of each variable was explored by Fisher’s criterion. Two machine learning models were established to allow adaptation to more situations. Bagging Tree, Gaussian process, support vector machine (SVM), and decision tree were used for prediction.
K2 after one year of orthokeratology (K2_after) was most important in the prediction of K1 and K2. Bagging Tree performed best in both models 1 and 2 for K1 prediction (R = 0.812, RMSE = 0.855 in model 1 and R = 0.812, RMSE = 0.858 in model 2) and K2 prediction (R = 0.831, RMSE = 0.898 in model 1 and R = 0.837, RMSE = 0.888 in model 2). In model 1, the difference was 0.006 ± 1.34 D (p = 0.93) between the predictive value of K1 and the true value of K1 (K1_before) and was 0.005 ± 1.51 D(p = 0.94) between the predictive value of K2 and the true value of K2 (K2_before). In model 2, the difference was −0.056 ± 1.75 D (p = 0.59) between the predictive value of K1 and K1_before and was 0.017 ± 2.01 D(p = 0.88) between the predictive value of K2 and K2_before.
Bagging Tree performed best in predicting K1 and K2. Machine learning can be applied to predict the corneal curvature for those who cannot provide the initial corneal parameters in the outpatient clinic, providing a relatively certain degree of reference for the refitting of the Ortho-k lenses.
Biometric factors and orthokeratology lens parameters can influence the treatment zone diameter on corneal topography in Corneal Refractive Therapy lens wearers
2023, Contact Lens and Anterior Eye
Citation Excerpt :
However, previous studies on the factors that affect the treatment zone focused mainly on ocular aberrations and visual performance [15,25], and few studies have focused on which factors may influence the treatment area size for ortho-K lenses. The major parameters of the Paragon CRT lens, which is a 3-curve design, are named BC, RZD, and LZA [26]. The design concept of CRT is “proximity control”, equal to “adjacent partition control”, which means that the three adjacent partitions of the shaping lens are independent and the parameters are not linked.
To investigate the relationship between patients’ baseline biometric factors or lens parameters and the diameter of the treatment zone in young myopic children undergoing Corneal Refractive Therapy.
The data of patients undergoing Corneal Refractive Therapy lens treatment within two years were retrospectively reviewed. Baseline clinical data, including sex, age, refractive power, corneal topography readings, ocular optical biometric measurements, and Corneal Refractive Therapy lens parameters, were subjected to Pearson, Spearman, and partial correlation analyses to identify the potential factors that may influence treatment zone diameter on corneal topography. Logistic and linear regression analyses were used to predict the treatment zone size.
The Right eyes of 309 patients were included in this study. The spherical refraction, flat keratometric reading, Reverse Zone Depth 2, Landing Zone Angle 1, and lens diameter were independent factors of treatment zone diameter. In the multivariate analyses, Landing Zone Angle 1 was positively correlated, while Reverse Zone Depth 2 and lens diameter were negatively correlated with the size of the treatment area. The accuracy of logistic regression in predicting the treatment zone size was 71.5%.
Adjustments to Corneal Refractive Therapy lens parameters may influence the treatment zone diameter on corneal topography. A higher Reverse Zone Depth 2, smaller Landing Zone Angle 1, and larger lens diameter can lead to a smaller treatment zone for Corneal Refractive Therapy lens treatment.
Machine learning algorithm improves accuracy of ortho-K lens fitting in vision shaping treatment
2022, Contact Lens and Anterior Eye
Citation Excerpt :
Today, machine learning and artificial intelligence (AI) have provided very powerful tools for clinical work in optometry [19]. Previous studies have proposed the use of machine learning to establish AI models for corneal refractive therapy (CRT) ortho-k lenses [20]. The purpose of this research is to propose a high-precision and convenient machine learning model for VST orthokeratology lens selection that can reduce the number of lens trials in the fitting process and make lens orders faster, safer, and more accurate, with an improvement over a previously published calculation method [21].
To construct a machine learning (ML)-based model for estimating the alignment curve (AC) curvature in orthokeratology lens fitting for vision shaping treatment (VST), which can minimize the number of lens trials, improving efficiency while maintaining accuracy, with regards to its improvement over a previous calculation method.
Data were retrospectively collected from the clinical case files of 1271 myopic subjects (1271 right eyes). The AC curvatures calculated with a previously published algorithm were used as the target data sets. Four kinds of machine learning algorithms were implemented in the experimental analyses to predict the targeted AC curvatures: robust linear regression models, support vector machine (SVM) regression models with linear kernel functions, bagging decision trees, and Gaussian processes. The previously published calculation method and the novel machine learning method were then compared to assess the final parameters of ordered lenses.
The linear SVM and Gaussian process machine learning models achieved the best performance. The input variables included sex, age, horizontal visible iris diameter (HVID), spherical refraction (SER), cylindrical refraction, eccentricity value (e value), flat K (K1) and steep K (K2) readings, anterior chamber depth (ACD), and axial length (AL). The R-squared values for the output AC1K1, AC1K2 and AC2K1 values were 0.91, 0.84, and 0.73, respectively. The previous calculation method and machine learning methods displayed excellent consistency, and the proposed methods performed best based on flat K reading and e values.
The ML model can provide practitioners with an efficient method for estimating the AC curvatures of VST lenses and reducing the probability of cross-infection originating from trial lenses, which is especially useful during pandemics, such as that for COVID-19.
Development of a Machine-Learning–Based Tool for Overnight Orthokeratology Lens Fitting
2024, Translational Vision Science and Technology
Research progress of artificial intelligence in the prevention and control of myopia
2023, International Eye Science

View all citing articles on Scopus

¹: These two authors contributed equally.

View full text

Machine learning based strategy surpasses the traditional method for selecting the first trial Lens parameters for corneal refractive therapy in Chinese adolescents with myopia

Abstract

Purpose

Methods

Results

Conclusions

Introduction

Section snippets

Clinical characteristics

Estimated RZD and LZA values in the four quadrants

Differences in corneal morphology between chinese and western patients

Conclusion

Funding support

Declaration of Competing Interest

Acknowledgements

Prog Retin Eye Res

Ophthalmology

Contact Lens Anterior Eye

Curr Eye Res

Ophthalmology

Cornea

Prog Retin Eye Res

Am J Ophthalmol

Ophthalmology

Myopia: an epidemic of possibilities?

Optom Vis Sci

Refractive errors in an elderly chinese population in Taiwan: the shihpai eye study

Invest Ophthalmol Vis Sci

The myopia boom

Nature

Myopia control using toric orthokeratology (TO-SEE study)

Invest Ophthalmol Vis Sci

Orthokeratology to control myopia progression: a meta-analysis

PLoS One

Long-term effect of overnight orthokeratology on axial length elongation in childhood myopia: a 5-year follow-up study

Invest Ophthalmol Vis Sci

Overnight orthokeratology - response

Optom Vis Sci

Orthokeratology review and update

Clin Exp Optom

Corneal modeling

Cornea