Development of a machine learning model for predicting myopia progression in children wearing peripheral defocus spectacle lenses

Background. Myopia remains one of the leading causes of visual impairment worldwide. A wide range of therapeutic and preventive strategies is currently applied in clinical practice to slow myopia progression. In recent years, increasing attention has been directed toward the use of artificial intelligence (AI) for predicting disease progression and treatment outcomes. Machine learning techniques enable the development of predictive models based on baseline clinical data, thereby improving individualized treatment planning. Purpose: To develop and validate a machine learning model for predicting myopia progression at the 12-month follow-up in children wearing peripheral defocus spectacle lenses. Materials and methods. A dataset comprising 48 eyes of 48 pediatric patients fitted with peripheral defocus spectacle lenses was analyzed. Binary classification models were developed using Python 3 and the scikit-learn, XGBoost, and LightGBM libraries. Eight machine learning algorithms were evaluated: XGBoost, Random Forest, Gradient Boosting, LightGBM, Extra Trees, Logistic Regression, Decision Tree, and K-Nearest Neighbors. The outcome variable was defined as a binary indicator of treatment effectiveness: favorable outcome (24 eyes) – annual myopia progression less than −1.00 diopters at 12 months; unfavorable outcome (24 eyes) – progression of −1.00 diopters or greater. The dataset was divided into training and test sets in a 67:33 ratio. Feature importance was assessed using built-in feature importance methods. Results. The highest predictive performance on the test dataset was demonstrated by the XGBoost model, with a ROC AUC of 0.906. The model achieved an accuracy of 0.875, sensitivity of 0.875, and specificity of 0.875. The most influential predictors of treatment outcome were: minimum keratometry (Kmin) at baseline (0.382), baseline spherical equivalent refraction (0.217), baseline corneal radius of curvature (0.184), baseline axial length (0.102), patient age (0.064), and maximum keratometry (Kmax) at baseline (0.051). Conclusion. The proposed machine learning model demonstrated excellent predictive performance for forecasting the outcome of peripheral defocus spectacle lens wear in children at the 12-month follow-up. A clinical decision-support calculator based on this model has been developed for practical application.

1. Fricke TR, Jong M, Naidoo KS, Sankaridurg P, Naduvilath TJ, Ho SM, Wong TY, Resnikoff S. Global prevalence of visual impairment associated with myopic macular degeneration and temporal trends from 2000 through 2050: systematic review, meta-analysis and modelling. Br J Ophthalmol. 2018 Jul;102(7):855–862. doi: 10.1136/bjophthalmol-2017-311266

2. Holden BA, Fricke TR, Wilson DA, Jong M, Naidoo KS, Sankaridurg P, Wong TY, Naduvilath TJ, Resnikoff S. Global Prevalence of Myopia and High Myopia and Temporal Trends from 2000 through 2050. Ophthalmology. 2016 May;123(5):1036– 1042. doi: 10.1016/j.ophtha.2016.01.006

3. Libman ES, Shakhova EV. Blindness and disability due to pathology of the organ of vision in Russia // Bulletin of ophthalmology. 2006(1):78–79. (In Russ.)

4. Avetisov ES, Tarutta EP. The three-factor theory of the origin of myopia and its practical significance. Topical issues of ophthalmology: tr. scientific-practical conf., dedicated. in memory of Hermann von Helmholtz. M., 1995:101. (In Russ.)

5. Milash SV, Epishina MV, Toloraya RR. Modern optical me thods of peripheral defocus correction. Russian Ophthalmological Journal. 2019;12(4):92–98. (In Russ.) doi: 10.21516/2072-0076-2019-12-4-92-98

6. Walker TW, Mutti DO. The effect of accommodation on ocular shape. Optom Vis Sci. 2002 Jul;79(7):424–430. doi: 10.1097/00006324-200207000-00010

7. He JC, Gwiazda J, Thorn F, Held R, Vera-Diaz FA. The association of wavefront aberration and accommodative lag in myopes. Vision Res. 2005 Feb;45(3):285–290. doi: 10.1016/j.visres.2004.08.027. Erratum in: Vision Res. 2005 Aug;45(18):2470.

8. Sreenivas an V, Irving EL, Bobier WR. Can current models of accommodation and vergence predict accommodative behavior in myopic children? Vision Res. 2014 Aug;101:51–61. doi: 10.1016/j.visres.2014.05.008

9. Hung GK, Ciuffreda KJ. An incremental retinal defocus theory of the development of myopia. Comments on Theoretical Biology. 2003(8):511–538. doi: 10.1080/08948550302433

10. Hung GK, Ciuffreda KJ. Differential retinal-defocus magnitude during eye growth provides the appropriate direction signal. Med Sci Monit. 2000;6(4):791–795.

11. Hung GK, Ciuffreda KJ. Incremental retinal defocus theory predicts experimental effect of under-correction on myopic progression. JBO. 2004(3):59–63.

12. Rose KA, Morgan IG, Ip J, Kifley A, Huynh S, Smith W, Mitchell P. Outdoor activity reduces the prevalence of myopia in children. Ophthalmology. 2008 Aug;115(8):1279–1285. doi: 10.1016/j.ophtha.2007.12.019

13. Papadogiannis P, Börjeson C, Lundström L. Comparison of optical myopia control interventions: effect on peripheral image quality and vision. Biomed Opt Express. 2023 Jun 6;14(7):3125–3137. doi: 10.1364/BOE.486555

14. Lam CS, Tang WC, Lee PH, Zhang HY, Qi H, Hasegawa K, et al. Myopia control effect of defocus incorporated multiple segments (DIMS) spectacle lens in Chinese children: results of a 3-year follow-up study. Br J Ophthalmol. 2022;106:1110–1114.

15. Lam CSY, Tang WC, Tse DY, Lee RPK, Chun RKM, Hasegawa K, et al. Defocus Incorporated Multiple Segments (DIMS) spectacle lenses slow myopia progression: a 2-year randomised clinical trial. Br J Ophthalmol. 2020;104:363–368.

16. Bao J, Huang Y, Li X, Yang A, Zhou F, Wu J, et al. Spectacle lenses with aspherical lenslets for myopia control vs singlevision spectacle lenses: a randomized clinical trial. JAMA Ophthalmol. 2022;140:472–478.

17. Bao J, Yang A, Huang Y, Li X, Pan Y, Ding C, et al. One-year myopia control efficacy of spectacle lenses with aspherical lenslets. Br J Ophthalmol. 2022;106.

18. Gantes-Nunez J, Jaskulski M, Lopez-Gil N, Kollbaum PS. Optical characterisation of two novel myopia control spectacle lenses. Ophthalmic Physiol Opt. 2023;43:388–401.

19. Jaskulski M, Singh NK, Bradley A, Kollbaum PS. Optical and imaging properties of a novel multi-segment spectacle lens designed to slow myopia progression. Ophthalmic Physiol Opt. 2020;40:549–56.

20. Sankaridurg P, Weng R, Tran H, Spiegel DP, Drobe B, Ha T, et al. Spectacle lenses with highly aspherical lenslets for slowing myopia: a randomised, double-blind, cross-over clinical trial. Am J Ophthalmol. 2022;247:18–24.

21. Guo H, Li X, Zhang X, Wang H, Li J. Comparing the effects of highly aspherical lenslets versus defocus incorporated multiple segment spectacle lenses on myopia control. Sci Rep. 2023.

22. Markova EY, Belikova TV, Kechin EV, Shikhalieva EA, Postolnik AA, Nikitina AE, Navruzalieva DY. Application of artificial intelligence in prevention of myopia progression. Solving the problem. Rossiyskaya detskaya oftalmologiya. 2024;3(49):54–60. (In Russ.) doi: https://doi.org/10.25276/2307-6658-2024-3-54-60

23. ROC-AUC: this is a metric for assessing the quality of a machine learning model. (In Russ.) URL: https://sky.pro/wiki/analytics/roc-auc-eto-metrika-otsenki-kachestva-modeli-mashinnogo-obucheniya/

24. Avetisov SE. The optical system and refraction of the eye. Eye diseases. Moscow: Ophthalmology; 2015:85-112. (In Russ.)

25. Kechin EV, Shikhalieva EA, Kostenev SV. Calculator for predicting myopia progression in children using perifocal glasses. Certificate of state registration of a computer program №b2025691882 dated 18.11.2025. (In Russ.)

26. Tong HJ, Huang ZM, Li YL, Chen YM, Tian B, Ding LL, Zhu LL. Machine learning to analyze the factors influencing myopia in students of different school periods. Front Public Health. 2023;11:1169128. doi: 10.3389/fpubh.2023.1169128