As part of the panel discussion, the importance of examining the evidence supporting the use of artificial intelligence tools was highlighted. Optometry 2049: AI and the future of vision care, Tomorrow’s Optometry (14-15 June, Harrogate Convention Center).
Michael Holler, a consultant optometrist at medical technology company Cascader, told attendees that AI offers “potentially huge benefits” to patients.
“Ophthalmology is the NHS’s largest service, with more than 11 million appointments each year,” he said.
He pointed out that although one in 10 NHS outpatient appointments are eye-related, there are significant waiting times for many patients to be seen.
Holler highlighted the potential for AI to improve the accuracy of referrals to hospital ophthalmology services and support local surveillance services.
“This frees up space in the NHS for patients who urgently need to be seen,” he says.
Horler noted that some innovative AI products have failed to gain traction due to poor usability.
“If you have something that takes 14 clicks to work, it’s going to fail because people won’t use it,” he observed.
He emphasized the importance of educating practitioners about what factors to consider when considering investments in AI tools, such as whether the technology has been independently validated and details about the population it was trained on.
“People need to be aware of all of these things so that when they go to a trade show and see a fancy stand with a device on it, they can critically evaluate it and decide if it’s actually the right thing to use,” Holler said.
Reflecting on who is responsible when decisions are supported by AI, Holler said he envisions AI tools constantly monitoring humans and monitoring decisions recommended by AI.
“Accountability is a thorny issue,” he said.
Regional optometrist Aidan Hussain pointed out that many of the approved applications of AI in eye care are focused on hospital settings.
“Despite a lot of smart work being done in eye care, these tools are not necessarily designed to be used in primary care. In terms of visible, day-to-day use, we are still really waiting for this technology to impact us,” he reflected.
Like Horler, Hussain emphasized the value of keeping clinicians up to date on AI.
“In primary care, if my signature is on a prescription, I am 100% responsible,” he says.
“Recognizing the limitations of AI, whether it is essentially a black box, how it was trained, how it came to its decisions, is important when my name is on the record card,” Hussain said.
Professor Alicia Rudnitska from City St George’s, University of London, revealed to participants that a commercial company is involved in research evaluating algorithms used to identify diabetic eye disease from retinal images.
He pointed out that the NHS diabetic eye screening program receives around 3.5 million appointments each year.
Rudnitska pointed out that the number of people eligible to be screened for diabetic eye disease in the UK is increasing.
“AI offers the opportunity to triage patients whose risk is low enough that human grading is not required,” she said.
The professor of statistical epidemiology emphasized the importance of evaluating AI tools before introducing them into communities.
“No algorithm is perfect,” Rudnitska said.
“What’s really important for these AI tools is that they work as intended in the environments they are deployed in, and that we evaluate them to work equally well for all tools,” she said.
Rudnicka observed that there is a wide range of AI tools, from low-risk tools such as AI Scribe to high-risk tools aimed at identifying specific eye diseases and monitoring their progression.
She shared that optometrists do not suddenly need to become computer scientists to harness the potential of AI.
“The important thing is to know how to interact with this tool, to know how it works and the range of its operation,” Rudnitska said.
Dr. Wen Hwa Lee, Chief Executive of Foresight Research, reflected on the limitations of AI and shared an example of an algorithm trained to detect skin cancer.
“The dataset used for training was a mostly Caucasian population, so performance degraded when deployed to non-Caucasian skin,” Lee explained.
In response to concerns about whether companies are profiting from patient data, Lee pointed out that people with smartphones are already susceptible to this effect.
“You’re already monetized for things that don’t necessarily serve you,” he said.
Lee shared that healthcare professionals need to consider the benefits of sharing data for innovation when discussing data with colleagues and patients.
“Every drug we have today was developed with the help of someone who very kindly donated their data. I welcome that, but it’s up to you as an individual,” he said.
