In The Pipeline

Artificial Intelligence and Fundus Photography Can Predict High Myopia

October 17, 2022

By Li Lian Foo, MD, MMed (Ophth), FRCOphth, BEng (1st class honors)

With the recent development of several enterprise-grade ophthalmic deep learning systems, AI has proven useful in disease classification and decision-making support. AI could also be applied similarly to aid myopia management.

Over the past decade, myopia has risen rapidly in prominence due to rising awareness among health care authorities and parents. Technological advancements and rapidly changing lifestyles worldwide are key drivers contributing to a global trend of accelerating myopia prevalence. If left unchecked, it is foreseeable that future generations will be left with an unimaginable disease burden to bear. In response, a palpable sense of urgency has taken hold among authorities and parents alike to deliver a robust solution to stem this tide. 

These fundal images represent classic pathologies directly related to high myopia that could lead to irreversible blindness. Left: myopic macular degeneration; Right: lacquer crack with subretinal bleed

Apart from being one of the leading causes of uncorrected, reversible visual impairment in the world,1-3 myopia can also cause potentially irreversible and debilitating disease burden during adulthood. Individuals with high myopia are at a higher risk of developing sight-threatening complications such as retinal detachment, myopic macular degeneration, open-angle glaucoma, and choroidal neovascularization.5,6 Furthermore, retinal degenerations such as chorioretinal atrophy and lacquer cracks have no known treatments to date. Apart from increasing the public health care burden,7 these complications would also have profound societal cost due to the loss of individuals’ productivity.8

A New Way to Identify Children at Risk of High Myopia
Another troubling statistic is the rising proportion of highly myopic patients among the young.4 To reverse this course, it is imperative to readily identify children who are at higher risk of developing high myopia at an early stage. Once identified, timely, effective treatment can be commenced so that myopic progression can be retarded for these individuals. 

Available interventions such as atropine eye drops and optical devices (specialized myopia control spectacle lenses and contact lenses) have proven to be effective in reducing myopic progression in children.9 However, what is currently lacking is the ability to screen and deliver the appropriate treatment to targeted individuals who are at an early stage of myopia. This is a key area of research where we seek to identify at-risk children with greater precision based on simple, accessible, and objective measures.

Currently, the known risk factors of high myopia include younger age of myopia onset,12-14 higher myopia diagnosed at presentation,15 rapid myopia progression,16,17 reduced outdoor activities and increased near work time,16 parental myopia,16 educational years,18 and polygenic risk scores.19,20 However, translating these factors into clinical practice can be challenging due to questionable information accuracy subject to recall bias, latency, and complicated testing procedures. As such, alternative strategies need to be explored.

Specifically, digital health care technology could be harnessed to develop adjunctive solutions that provide scalability, portability, and reliability to address the problem. Artificial intelligence (AI) could provide a solution to these needs. With the recent development of several enterprise-grade ophthalmic deep learning systems, AI has proven useful in disease classification and decision-making support.21,22 AI could also be applied similarly to aid myopia management. This would derive benefits such as increased automation and scalability, reduced operator dependence, and improved support for remote monitoring — preferred attributes for managing a high-prevalence disease such as myopia. 

How Our Team is Addressing the Need to Identify At-Risk Children
To address this, our team has been working on the development of a predictive algorithm that would address these challenges and provide the greatest clinical impact. First, we have a target age group of children between 6 to 12 years old who are most vulnerable to myopic progression and yet also amenable to myopia control therapies.23,24 Second, we used only objective inputs to avoid biases related to subjective recall. These are collected at a single time point (baseline) to eliminate the need for repeated, cumbersome longitudinal follow-up before a clinical decision can be made. This avoids the unnecessary delay of treatment for these high-risk individuals. 

Preliminary performance of the AI algorithm is promising, based on the Area-under-curve (AUC), sensitivity, and specificity indices. Moreover, fundus imaging heatmaps generated from our deep learning models identified the disc and macular as areas of interest, consistent with areas of myopic disc changes and myopic macular degeneration. This could suggest that our algorithm can detect subtle features of high myopia and axial length elongation that were not apparent clinically among these high-risk children. Currently, our algorithm is developed using a longitudinal school-based dataset and would require further testing in other cohorts with different study populations. We are currently pursuing external validation involving children of different ethnicities and geographical locations.

Presently, a model of care that provides early identification, continuous surveillance, and individualized treatment would require a significant amount of resources and manpower to fulfill. This model of care is likely cost-ineffective, which hampers the sustainability of such efforts. This impending challenge should be met with appropriate pre-emptive actions. There is hence a pressing need to transform clinical practice and health care policies in myopia management. Using a single baseline fundus image to identify at-risk schoolchildren is highly promising due to the ease of operational implementation. With the advancement in digital imaging technology and increasing availability of said equipment, fundus imagery could be a valuable clinical predictive tool. Through early identification, targeted and timely myopia control therapies may be instituted to reduce the risk of developing high myopia in young ones and for generations to come. 

 

Dr. Li Lian Foo is a Consultant Ophthalmologist, Refractive Surgery Department and Myopia Centre with the Singapore National Eye Centre and Clinical Assistant Professor with Duke-NUS Medical School.

 

References

  1. Pararajasegaram R. VISION 2020-the right to sight: from strategies to action. American journal of ophthalmology 1999; 128(3): 359-60.
  2. Holden BA, Wilson DA, Jong M, et al. Myopia: a growing global problem with sight-threatening complications. Community eye health 2015; 28(90): 35.
  3. Blindness GBD, Vision Impairment C, Vision Loss Expert Group of the Global Burden of Disease S. Trends in prevalence of blindness and distance and near vision impairment over 30 years: an analysis for the Global Burden of Disease Study. The Lancet Global health 2021; 9(2): e130-e43.
  4. WHO. The impact of myopia and high myopia. 2015. https://www.who.int/blindness/causes/MyopiaReportforWeb.pdf.
  5. Wong TY, Ferreira A, Hughes R, Carter G, Mitchell P. Epidemiology and disease burden of pathologic myopia and myopic choroidal neovascularization: an evidence-based systematic review. American journal of ophthalmology 2014; 157(1): 9-25 e12.
  6. Ikuno Y. Overview of the Complications of High Myopia. Retina 2017; 37(12): 2347-51.
  7. Morgan IG, Ohno-Matsui K, Saw SM. Myopia. Lancet 2012; 379(9827): 1739-48.
  8. Naidoo KS, Fricke TR, Frick KD, et al. Potential Lost Productivity Resulting from the Global Burden of Myopia: Systematic Review, Meta-analysis, and Modeling. Ophthalmology 2019; 126(3): 338-46.
  9. Ang M, Flanagan JL, Wong CW, et al. Review: Myopia control strategies recommendations from the 2018 WHO/IAPB/BHVI Meeting on Myopia. Br J Ophthalmol 2020; 104(11): 1482-7.
  10. Liu YM, Xie P. The Safety of Orthokeratology–A Systematic Review. Eye & contact lens 2016; 42(1): 35-42.
  11. Gong Q, Janowski M, Luo M, et al. Efficacy and Adverse Effects of Atropine in Childhood Myopia: A Meta-analysis. JAMA ophthalmology 2017; 135(6): 624-30.
  12. Chua SY, Sabanayagam C, Cheung YB, et al. Age of onset of myopia predicts risk of high myopia in later childhood in myopic Singapore children. Ophthalmic & physiological optics : the journal of the British College of Ophthalmic Opticians 2016; 36(4): 388-94.
  13. Hu Y, Ding X, Guo X, Chen Y, Zhang J, He M. Association of Age at Myopia Onset With Risk of High Myopia in Adulthood in a 12-Year Follow-up of a Chinese Cohort. JAMA ophthalmology 2020; 138(11): 1129-34.
  14. Jensen H. Myopia in teenagers. An eight-year follow-up study on myopia progression and risk factors. Acta ophthalmologica Scandinavica 1995; 73(5): 389-93.
  15. Gwiazda J, Hyman L, Dong LM, et al. Factors associated with high myopia after 7 years of follow-up in the Correction of Myopia Evaluation Trial (COMET) Cohort. Ophthalmic epidemiology 2007; 14(4): 230-7.
  16. Parssinen O, Kauppinen M. Risk factors for high myopia: a 22-year follow-up study from childhood to adulthood. Acta ophthalmologica 2019; 97(5): 510-8.
  17. Lanca C, Foo LL, Ang M, et al. Rapid Myopic Progression in Childhood Is Associated With Teenage High Myopia. Invest Ophthalmol Vis Sci 2021; 62(4): 17.
  18. Liu L, Jiang D, Li C, et al. Relationship between Myopia Progression and School Entrance Age: A 2.5-Year Longitudinal Study. Journal of ophthalmology 2021; 2021: 7430576.
  19. Lanca C, Kassam I, Patasova K, et al. New Polygenic Risk Score to Predict High Myopia in Singapore Chinese Children. Translational vision science & technology 2021; 10(8): 26.
  20. Ghorbani Mojarrad N, Plotnikov D, Williams C, Guggenheim JA, Eye UKB, Vision C. Association Between Polygenic Risk Score and Risk of Myopia. JAMA ophthalmology 2020; 138(1): 7-13.
  21. Ting DSW, Cheung CY, Lim G, et al. Development and Validation of a Deep Learning System for Diabetic Retinopathy and Related Eye Diseases Using Retinal Images From Multiethnic Populations With Diabetes. Jama 2017; 318(22): 2211-23.
  22. Milea D, Najjar RP, Zhubo J, et al. Artificial Intelligence to Detect Papilledema from Ocular Fundus Photographs. The New England journal of medicine 2020; 382(18): 1687-95.
  23. Group C. Myopia stabilization and associated factors among participants in the Correction of Myopia Evaluation Trial (COMET). Investigative ophthalmology & visual science 2013; 54(13): 7871-84.
  24. Gifford KL, Richdale K, Kang P, et al. IMI – Clinical Management Guidelines Report. Investigative ophthalmology & visual science 2019; 60(3): M184-M203.
To Top