AI sees into the eye better than humans

The first part of the symposium was dedicated to artificial intelligence in diagnostics. Katherine Talcott from the Cole Eye Institute (Cleveland Clinic) described how AI is transforming retinal imaging. An examination using optical coherence tomography (OCT) — essentially an ultrasound using light that creates cross-sections of the retina at micrometer resolution — generates hundreds of scans. The physician must review them manually, and the risk of overlooking a subtle abnormality is real.

AI is already helping, primarily in screening for diabetic retinopathy, which is retinal damage caused by diabetes. The LumineticsCore platform (formerly IDx-DR) can autonomously decide from a fundus image whether a patient needs referral to a specialist — without physician intervention. It is the first AI system ever approved by the FDA for autonomous diagnosis in medicine. The problem is that similar autonomous screening tools have not yet been widely deployed in the Czech Republic.

Talcott also mentioned CIRRUS PathFinder from Zeiss — a platform that automatically flags abnormal findings from OCT scans and recommends when the patient should be referred to a specialist. For the Czech reality, it is noteworthy that Zeiss has a strong presence in the country and similar technologies are gradually making their way into domestic hospitals.

Robot in the eye: Cataract as the ideal candidate

Why exactly is cataract surgery the ideal candidate for robotics? Vance Thompson summed it up clearly: compared to general surgery, ocular anatomy has extremely low variability. The eye is a small, consistent space, the procedure is routine, standardized, and performed in high volumes — approximately 30 million cataract surgeries are performed worldwide each year.

What specifically do robots offer in eye surgery? According to Thompson, the benefits include:

• Shorter training time — the surgeon can learn on a simulator and the robot compensates for a lack of dexterity
• Improved ergonomics — the surgeon does not sit hunched over a microscope for hours
• More precise incisions and reduced tremor — the robot filters out the natural hand tremor
• Better refractive outcomes — more precise lens placement means sharper vision after surgery

Yu-Hsuan Huang from Taiwan presented specific robotic systems that already exist. Preceyes from Zeiss is a teleoperated platform where the surgeon controls instruments inside the eye via a robotic arm. Polaris from Horizon Surgical Systems was originally designed for partially automated cataract surgery with two robotic arms — but in its first human trial, it ended up as "just" a teleoperated platform without autonomy. This shows just how wide the gap is between a research concept and clinical reality.

The Israeli startup ForSight Robotics with its ORYOM system goes even further — the surgeon sits at a console with 3D visualization and operates remotely using handheld controllers. The system uses a bimanual approach, meaning a two-handed technique similar to what surgeons use today.

A Czech footprint: Pavel Stodulka and Europe leading robotic surgery

For Czech readers, the key name is Pavel Stodulka, head physician of the Gemini Eye Clinic in Zlín, who spoke at the congress with a European perspective. Stodulka, one of the most experienced eye surgeons in the Czech Republic (having performed over 100,000 operations), spoke about how robotic cataract surgery is coming, whether we want it or not. According to him, Europe needs to collect real-world clinical data in order to properly evaluate this technology.

Stodulka also introduced the FemtoMatrix platform from the French company Keranova. It is a femtosecond laser that uses OCT guidance for energy optimization — it can cut to various depths with minimal energy, meaning less damage to surrounding tissue. Stodulka is himself involved in the development of this laser.

Stodulka also mentioned five levels of automation in eye surgery: from manual surgery through femtosecond laser (FLACS), robotic positioning, instrument manipulation, up to so-called supervised autonomy — a state in which the robot performs parts of the procedure on its own under the surgeon's supervision. "That is the future," Stodulka said, "and real-world data is essential to get there."

Simulators: The end of learning on patients

Simulation formed an important part of the symposium. Anshu Arundhati from the Singapore National Eye Centre put it bluntly: "Learning at the patient's expense is long past being the way forward." The COVID-19 pandemic permanently accelerated the adoption of digital learning platforms, and eye centers around the world are now deploying VR simulators as standard.

Key tools include Eyesi Surgical from Haag-Streit — a simulator for cataract and vitreoretinal surgery. At the Singapore National Eye Centre, completing the Eyesi module is a mandatory requirement before a resident even sets foot in the operating room. The simulator also offers a slit lamp for practicing basic examinations.

Another tool is VisioGen from Zeiss, which uses machine learning to analyze surgical videos. It can compare a resident's performance against a database of thousands of expert surgeries and automatically assess instrument path, speed of movement, and the frequency of inefficient maneuvers.

David Cherwek from Orbis International emphasized that simulators do not turn a novice into a surgeon — they make them "ready for the OR." According to him, the problem is not the technology but its adoption: "The US is paradoxically lagging behind in this." By contrast, Europe is leading by example thanks to initiatives like ESCRS iLearn and traveling simulators (Eyesi on wheels moving between European cities).

AI calculates artificial lenses more accurately than mathematical formulas

Woong-Joo Whang from the Catholic University of Korea presented a study showing why AI outperforms traditional calculations in surgical planning. When calculating the so-called effective lens position (ELP) — i.e. exactly where the artificial intraocular lens will end up inside the eye — classical mathematical formulas fail. The problem is that the significance of individual parameters (eye length, corneal curvature, anterior chamber depth) changes depending on the size of the eye.

Whang's team compared 35 deep learning models (30 of which used various attention mechanisms, which allow the model to focus on the most important relationships between parameters). The result: 20 models outperformed traditional regression analysis and 22 out of 30 attention-enhanced models were better than standard deep learning. In other words — AI can "focus" on what truly matters in a given eye.

An ethical note: When AI writes science

The symposium was closed by Joaquín Fernández from Spain with an unexpectedly philosophical contribution. He addressed the risks of AI in scientific publishing — and referenced Aristotle's five intellectual virtues. His warning is simple: if we let AI read studies, write papers, and summarize findings for us, we will lose episteme — deep knowledge of the field — as well as nous — the intuition that lets us recognize what is quality science and what is not.

Fernández reminded that the author must verify the accuracy, completeness, and impartiality of all AI-generated content — and above all ensure that the AI tool does not acquire rights to the input materials (i.e. that it does not use them for training). At a time when empirical data shows a growing number of scientific articles written with AI assistance, this is a message that applies well beyond ophthalmology.

What it means for the Czech Republic

The Czech Republic holds a strong position in this field. Pavel Stodulka is an internationally recognized expert who actively collaborates on the development of new technologies. Czech eye clinics such as Gemini and Lexum are among the European leaders and perform tens of thousands of surgeries each year. The question remains: when will the first fully robotic cataract surgery reach our country?

Based on the current pace of development and regulatory requirements, the first clinical deployment of robotic platforms in Europe can be expected within the next 3–5 years. Whether the Czech Republic will be among the first countries depends also on investment in equipment and the willingness of insurance companies to cover such procedures.