The Utilization of Artificial Intelligence in Myopia Management

Artificial intelligence (AI) evokes a range of responses, from amazement and intrigue to apprehension and insecurity. Whatever the sentiment, AI remains a permanent part of our future and is destined to improve our capacity to interpret and utilize information. In eyecare, this innovation arrives at a pivotal moment.

Myopia affected an estimated 28% of the world population in 2010. By 2050, this figure is predicted to increase to nearly 50%, with 10% of myopic individuals expected to develop high myopia (-5.00D or more).¹

Beyond the financial burden on individuals and healthcare systems, high myopia also raises the risk of serious eye complications, including retinal detachment, glaucoma, and cataracts.² Therefore, it is imperative that we use all of the technology available, including AI, to optimize outcomes for myopic patients.

What is artificial intelligence?

AI is a computer science field that develops systems capable of performing tasks usually associated with human intelligence. Imagine an AI system that can grade diabetic retinopathy by following a set of programmed rules.

Machine learning (ML) is a subset of AI where systems improve by learning from data. Using the same example, a machine learning model could be trained on thousands of labeled fundus images. It would learn to associate specific characteristics with disease severity, but humans would still need to determine the relevant features.

Moreover, deep learning, a subset of machine learning, mimics brain neural networks with multiple layers to learn from raw data. It can analyze fundus images without predefined features, automatically detecting patterns linked to disease severity. Though highly accurate, it requires large data and computing power.

How AI is currently being implemented in eyecare

The IDx‑DR autonomous AI system (also known as LumineticsCore) is an FDA-cleared AI platform that detects diabetic retinopathy from fundus photographs and reports whether the signs are referable. As it is autonomous, such a tool could be used in primary-care offices to expand diabetic retinopathy screening beyond eye clinics, enabling earlier detection and referral.³

A similar application has been examined in the collaborative research project between Moorfields Eye Hospital and Google DeepMind. Their ongoing project investigates the use of AI in analyzing OCT scans to identify retinopathy. The project also goes beyond disease detection by offering reasoning behind its outcomes.⁴

The use of AI extends beyond imaging. Dora, a Euphonia UK Conformity

Assessed (UKCA)-marked AI clinical assistant, can interact directly with patients. At Buckinghamshire Healthcare NHS Trust (BHT), Dora contacted cataract patients 3 weeks post-operatively to collect symptom feedback and second-eye preferences, reducing follow-ups by 60% and freeing staff for complex cases.⁵

Utilizing AI in myopia care

Myopia often develops in childhood and requires management during this critical period when interventions can slow its progression and reduce long-term risks.

Detecting myopia early in children is challenging because many adapt to their symptoms. Caregivers and teachers are usually responsible, but this depends on their awareness and understanding of eye health.

While screening by eyecare specialists in schools and health centers would highlight those at risk of pathologic myopia, such coverage is unrealistic for a vast and growing population, particularly in low-resource settings.

There are three key steps in myopia management:

Detecting myopia with AI

Tong et al. used machine learning models to identify risk factors for myopia among students in primary, middle, and high schools. Their results suggest that while genetic factors are influential in younger students, behavioral factors (such as outdoor time and reading posture) gain importance in older students.⁶

Remote tools like the SVOne aberrometer measure refractive error without a full clinical examination, but may require cycloplegia in children. Additionally, fundus photography can provide a reliable and noninvasive method of screening.

High myopia is associated with quantifiable biomarkers, such as:⁷

Denser fundus tessellation
Larger and tilted optic discs
Peripapillary atrophy

A study of over 85,000 fundus images from youths aged 6 to 18 was conducted, with key abnormalities labelled. Not only did the system achieve high accuracy (area under the curves [AUCs] > 0.93), but its sensitivity and specificity were comparable to those of human experts.⁸

If developed for real-world screening, fundus photography with AI analysis could allow earlier detection of myopia outside of clinics with minimal training. Mobile fundus cameras such as the ZEISS Visuscout 100 or Volk Instaview can provide high-quality digital images for AI processing.^9,10

Improving myopia prediction

Myopia, affected by both genetic and lifestyle factors, provides extensive data for prediction. AI systems, trained on specific markers, can offer accurate predictions reliably and efficiently, supported by vast processing power.

Data from the following sources have been used to develop AI systems for detecting myopia:

Electronic health records: A 2018 study developed a machine-learning model capable of predicting the onset of high myopia in children up to 10 years in advance.
- Trained on over 680,000 records, this was among the first studies to show that large datasets combined with AI processing can produce accurate forecasts of myopia progression.¹¹
Biometric and behavioral data: Another study incorporated ocular measurements, lifestyle habits, and environmental factors into their system to accurately predict adolescent myopia using machine learning.
- This method was able to distinguish between protective and risk factors, as well as identify modifiable risks to facilitate targeted interventions.¹²
Fundus imaging: The same fundus findings used for myopia detection can be monitored to predict progression. However, the method's accuracy is limited by the absence of environmental, behavioral, and refractive data.¹³

Leveraging AI to guide myopia management

Myopia management often involves a long-term plan, regularly adapted and reviewed as the child grows.

Common approaches include:

Spectacles and soft contact lenses designed to slow axial length elongation.
Atropine in low doses has been shown to slow down myopia progression by inhibiting scleral stretching.¹⁴
Orthokeratology involves reshaping the cornea to correct refractive error through the nightly use of a specially designed contact lens.

Predictive AI models can assist clinicians in selecting the most effective treatment options for their patients. For example, a recent study developed a machine learning model to predict which myopic children would respond best to orthokeratology (ortho-K) treatment. This retrospective analysis included 119 children and evaluated numerous baseline ocular and lifestyle factors as potential predictors of treatment success.¹⁵

Using a logistic regression model with LASSO feature selection, the study identified several key factors associated with better ortho-K efficacy, such as:¹⁵

Age
Baseline axial length
Pupil diameter
Nightly lens-wear duration
Daily time outdoors
Near-work time
Corneal diameter (white-to-white)
Corneal curvature
Posterior corneal astigmatism

Ultimately, the model achieved excellent predictive performance (AUC ~0.95). Such an ML-assisted tool can help eyecare professionals personalize myopia control by forecasting which patients are likely to benefit most from ortho-K therapy.¹⁵

Parental involvement in myopia management is also essential for success. Predictive models that can demonstrate likely myopia progression can help parents understand the importance of compliance, whether that means encouraging more outdoor activity or ensuring consistent spectacle wear.

Challenges and limitations

Despite significant advancements, few AI systems have successfully transitioned from research to routine clinical practice.

The most common barriers to implementing AI systems into clinical practice are:

Medical liability and data security: There are concerns regarding accountability if an AI system makes an error or hallucination. Furthermore, care would need to be taken in how sensitive health data is stored and shared.¹⁶
The “black box: problem: This refers to the inability to view an AI algorithm beyond inputs and outputs, thus making mistakes or misdiagnoses drawn from AI processing difficult to detect.¹⁷
Infrastructure and resources: Reduced access to imaging devices, reliable internet, and data storage in developing countries hampers the widespread adoption of AI tools in healthcare.
- Furthermore, these tools require significant amounts of power and hardware, creating financial challenges even in well-resourced environments.¹⁸

Looking ahead: What’s next for AI in myopia?

Research into using AI to tackle the myopia epidemic continues to grow. New technologies like ChatMyopia,¹⁹ an AI agent designed to enhance patient education and satisfaction, showcase innovative ways in which AI can help families better understand the condition.

Smartphone-based screening and portable imaging devices could make myopia detection and treatment more accessible and efficient. These innovations could enable early detection in schools, primary care settings, and underserved regions.

Looking ahead, the integration of AI into myopia management will probably involve not just detection and prediction, but also personalized treatment planning and patient engagement. If these tools can overcome current barriers, they could help ensure that pathologic myopia is detected and managed effectively during a patient’s critical period of progression, thereby preserving vision and quality of life.

5 key takeaways

Myopia prevalence is expected to reach nearly 50% of the world’s population by 2050, with high myopia posing significant risks.
AI enables earlier detection of myopia through fundus image analysis.
Predictive AI models can forecast myopia progression using biometric, behavioral, and genetic data
AI can support myopia management, guiding therapeutic choices including spectacles, contact lenses, atropine, or orthokeratology.
Key challenges remain, including liability, the “black box” problem, and equitable access. However, AI tools are moving closer to routine clinical use.