Ophthalmology balances the promises and challenges of AI

(Image Credit: AdobeStock/Татьяна Максимова)

Artificial intelligence (AI) is not a newcomer to medicine. In fact, its roots in health care trace back several decades, with early mentions in medical literature in the mid-20th century. What has changed dramatically in recent years is the sophistication and potential impact of AI on clinical practice, particularly in ophthalmology.

AI encompasses software programs that mimic human cognitive functions such as learning, reasoning, and decision-making. Advances in machine learning and deep learning algorithms, powered by superior computational capabilities, have propelled AI from theoretical constructs to practical tools with the potential to transform patient care.¹

The eye as a window for AI applications

The eye presents a unique opportunity for AI integration. Its accessible anatomy and the predominance of image-based diagnostics make ophthalmology particularly amenable to AI solutions.

From retinal imaging to visual field analysis, vast amounts of data are generated that can be harnessed by AI algorithms to enhance diagnostic accuracy and efficiency.² AI applications in ophthalmology span a broad spectrum of diseases and interventions:

• Retinal diseases: AI algorithms have demonstrated high accuracy in detecting diabetic retinopathy, age-related macular degeneration, and retinopathy of prematurity from retinal images.³ These tools can help screen large populations, especially in medically underserved areas, reducing the burden on specialists and facilitating early intervention.
• Glaucoma: Machine learning models have been developed to predict the risk of glaucoma progression by analyzing optic nerve head images and visual field data.⁴ Such predictive analytics can aid in tailoring treatment plans and monitoring strategies for individual patients.
• Cataract surgery: AI has been applied to improve IOL power calculations, reducing refractive errors post surgery.⁵ By analyzing biometric data, AI algorithms can provide more precise measurements, enhancing surgical outcomes.
• Corneal diseases and refractive surgery: In corneal imaging, AI assists in detecting keratoconus and planning refractive procedures. The integration of AI into topography and tomography systems helps in interpreting complex data patterns that might be challenging for the human eye.⁶

The black box concern

Despite the promise, one of the most voiced concerns about AI in medicine is its “black box” nature.⁷ Clinicians are wary of algorithms that provide outputs without transparent reasoning processes. Trusting a diagnosis or treatment recommendation without understanding the underlying rationale challenges the principles of evidence-based medicine.⁸

This opacity raises several ethical and practical issues, including the following:

• Accountability: Who is responsible if an AI system errs? The clinician, the software developer, or the institution?
• Bias and fairness: AI algorithms trained on nonrepresentative data may perpetuate or exacerbate health disparities. Ensuring diversity in training data is crucial to developing equitable AI tools.⁹
• Patient autonomy and trust: Patients have the right to understand how decisions about their care are made. The inability to explain AI-generated recommendations can undermine patient trust and informed consent.

Navigating the challenges

To harness the benefits of AI while mitigating its drawbacks, some strategies can be employed, including the following:

• Explainable AI (XAI): Developing algorithms that provide interpretable results can bridge the gap between complex computations and clinical reasoning. XAI aims to make the decision-making process of AI systems transparent and understandable.^7,10
• Rigorous validation: AI tools must undergo thorough testing in real-world clinical settings, not just in controlled environments. Cross-institutional studies and external validations can help ensure that AI applications are generalizable and reliable.¹¹
• Continuous monitoring and updates: AI systems should be regularly updated with new data to maintain accuracy over time. Implementing feedback mechanisms allows for the correction of errors and adaptation to new clinical insights.³
• Ethical frameworks and guidelines: Establishing clear guidelines regarding the use of AI in clinical practice can address issues of responsibility, consent, and data privacy. Professional societies have a role in developing standards for AI integration.^12,13

Embracing the future

The evolution of AI is rapid, and its potential in ophthalmology is vast. Although concerns about the black box nature of AI are valid, they should not overshadow the significant benefits that AI can bring to patient care. Instead of fearing AI, ophthalmologists should engage with it, contributing to its development and ensuring that it serves as a valuable adjunct to clinical expertise.

As clinicians, the role is to critically assess AI tools, understand their limitations, and integrate them thoughtfully into practice. Collaboration among ophthalmologists, data scientists, and ethicists is essential to shape AI applications that are both effective and aligned with the values of patient-centered care.

The International AI in Ophthalmology Society (IAIOph) is experiencing remarkable growth, with more than 1400 professionals already part of our community. Our mission at IAIOph is to foster collaboration between health care professionals and computer scientists to advance the field of AI in ophthalmology.

In the words of a colleague, “AI can be a tool of great value in diagnosing and treating complex eye problems. Let’s embrace AI and work closely with it but remain critical about the solutions it brings.”¹⁴

AI holds immense promise for advancing ophthalmology. By navigating the challenges thoughtfully and proactively, we can unlock the full potential of AI, enhancing our ability to diagnose and treat our patients and ultimately improve their lives.

Polat Goktas, PhD

E: polat.goktas@ucd.ie

Goktas is a distinguished research fellow at University College Dublin in Ireland, where he specializes in human-computer interaction, data analytics, statistics, generative artificial intelligence (AI), machine learning, explainable AI, and ethics and privacy in health care. He is also the secretary-general of the organizing committee for the International AI in Ophthalmology Society.

Andrzej Grzybowski, MD, PhD, MBA, MAE

E: ae.grzybowski@gmail.com

Grzybowski is a professor of ophthalmology at the University of Warmia and Mazury in Olsztyn and head of the Institute for Research in Ophthalmology at the Foundation for Ophthalmology Development in Poznań, both in Poland. He is also head of the organizing committee of the International AI in Ophthalmology Society.

References

1. Jin K, Yuan L, Wu H, Grzybowski A, Ye J. Exploring large language model for next generation of artificial intelligence in ophthalmology. Front Med (Lausanne). 2023;10:1291404. doi:10.3389/fmed.2023.1291404

2. Li H, Cao J, Grzybowski A, Jin K, Lou L, Ye J. Diagnosing systemic disorders with AI algorithms based on ocular images. Healthcare (Basel). 2023;11(12):1739. doi:10.3390/healthcare11121739

3. Grzybowski A, Singhanetr P, Nanegrungsunk O, Ruamviboonsuk P. Artificial intelligence for diabetic retinopathy screening using color retinal photographs: from development to deployment. Ophthalmol Ther. 2023;12(3):1419-1437. doi:10.1007/s40123-023-00691-3

4. Xu Z, Xu J, Shi C, et al. Artificial intelligence for anterior segment diseases: a review of potential developments and clinical applications. Ophthalmol Ther. 2023;12(3):1439-1455. doi:10.1007/s40123-023-00690-4

5. Stopyra W, Voytsekhivskyy O, Grzybowski A. Accuracy of 20 intraocular lens power calculation formulas in medium-long eyes. Ophthalmol Ther. 2024;13(7):1893-1907. doi:10.1007/s40123-024-00954-7

6. Kang D, Wu H, Yuan L, Shi Y, Jin K, Grzybowski A. A beginner’s guide to artificial intelligence for ophthalmologists. Ophthalmol Ther. 2024;13(7):1841-1855. doi:10.1007/s40123-024-00958-3

7. Goktas P. Ethics, transparency, and explainability in generative AI decision-making systems: a comprehensive bibliometric study. J Decis Syst. Published online October 10, 2024. doi:10.1080/12460125.2024.2410042

8. Durán JM, Jongsma KR. Who is afraid of black box algorithms? on the epistemological and ethical basis of trust in medical AI. J Med Ethics. Published online March 18, 2021. doi:10.1136/medethics-2020-106820

9. Grzybowski A, Jin K, Wu H. Challenges of artificial intelligence in medicine and dermatology. Clin Dermatol. 2024;42(3):210-215. doi:10.1016/j.clindermatol.2023.12.013

10. Lambert WC, Lambert MW, Emamian MH, Woźniak M, Grzybowski A. Artificial intelligence and the scientific method: how to cope with a complete oxymoron. Clin Dermatol. 2024;42(3):275-279. doi:10.1016/j.clindermatol.2023.12.021

11. Zhou Z, Zhang X, Tang X, Grzybowski A, Ye J, Lou L. Global research of artificial intelligence in strabismus: a bibliometric analysis. Front Med (Lausanne). 2023;10:1244007. doi:10.3389/fmed.2023.1244007

12. Lambert WC, Grzybowski A. Dermatology and artificial intelligence. Clin Dermatol. 2024;42(3):207-209. doi:10.1016/j.clindermatol.2023.12.014

13. Goktas P, Grzybowski A. Assessing the impact of ChatGPT in dermatology: a comprehensive rapid review. J Clin Med. 2024;13(19):5909. doi:10.3390/jcm13195909

14. Blanckaert J. The “black box” of artificial intelligence in ophthalmology. Ophthalmology Times Europe. October 7, 2024. Accessed January 6, 2025. https://europe.ophthalmologytimes.com/view/the-black-box-of-artificial-intelligence-in-ophthalmology