Fulfilling AI’s Promise

After the development of molecular biology in the 1950s and genomics in the 1980s, the current convergence of microelectronics and artificial intelligence (AI) marks the third great revolution in medical technology since the Flexner Report refocused medical education on science in the US in 1910, said Dimitri T Azar MD. However, while AI promises great strides in virtually every aspect of medicine, steps must be taken to ensure it does not introduce systemic biases, he added.

Multiple uses

In ophthalmology, actual and potential AI applications include analysis of corneal topography for ectasia signs, collagen cross-linking guidance, and cataract and refractive surgery planning, including topography analysis and IOL power calculation, Professor Azar said. Automated Fuchs’ dystrophy and glaucoma diagnosis and analysis of fundus photography and OCT for identifying retinal lesions, diabetic retinopathy, retinopathy of prematurity, and age-related macular degeneration exist or are in development.

Other potential AI applications include developing new dry eye diagnostic criteria, predicting Fuchs’ dystrophy progression, smart IOLs, guiding refractive surgery, and even robotic cataract surgery, Prof Azar said. And as miniaturization and AI converge, new, more economical diagnostic and monitoring devices will be available for clinical and home use. “There [will] be personalized refractive surgery based on knowledge that you [must] combine from many machines. You cannot rely on one device only.”

Regarding its role in ophthalmic education, he noted AI can supplement medical examination questions, provide multilingual translations, and create unique learning strategies. It has many roles in research, such as finding new treatments and targets, conducting literature reviews, drafting manuscripts, and recruiting patients.

In the office, AI can help synthesize diverse pieces of information to improve clinical management and patient consultations, taking notes and filing claims. “AI is going to be your virtual assistant,” Prof Azar said.

Addressing hurdles

Achieving the promise of AI requires addressing significant challenges, Prof Azar said. Bias in training data sets is one.

Because AI produces empirical models based on connections it finds among data points, its reliability depends on data that are accurate and representative of the target population. For example, an AI algorithm based primarily on eyes from European populations may not work as well or at all in an Asian population. Rather than try to correct bias in algorithms after their creation, Prof Azar said efforts should focus on ensuring data quality, representative sampling, and how algorithms interpret data.

And since AI models are constructed using self-generated rules, they present the question of how the user can know if the output is correct. Regulators grapple with AI’s reliability, and the guidelines they adopt to ensure it will heavily influence how and what kind of systems are developed.

Still, there is no reason to fear the unknown, Prof Azar said. “We need to ensure that AI will be a helpful tool, rather than a competitor, for early referrals and diagnosis.”

Prof Azar presented in the JCRS Symposium at the 2025 ASCRS annual meeting in Los Angeles.

Dimitri T Azar MD, MBA is CEO of LACRISTAT.AI and distinguished university professor, BA field chair of ophthalmological research, professor of ophthalmology, pharmacology, and bioengineering, and executive dean emeritus at the University of Illinois College of Medicine, Chicago, US. dazar@uic.edu