03 March 2026
Artificial intelligence (AI) models in practice were the subject of optometrist Kishan Devraj’s CPD lecture on the first morning of 100% Optical 2026 (Saturday 28 February).
Devraj began his Optical Academy presentation, entitled Artificial intelligence models in practice, by acknowledging that more people are now using AI on a daily basis than there were a handful of years ago.
He shared his own definition of AI, which he believes is “the ability of machines to perform tasks that typically require human intelligence – machines that can predict, based on input.”
Devraj noted that there are there are four layers involved in AI: the consumption of energy, computing infrastructure, chips (that is, the core hardware used), and the cloud.
“Many of you now use ‘the cloud’ on a daily basis. The cloud infrastructure layer allows us to build models on top of this,” Devraj said.
Devraj revealed that 230 million people globally ask health or wellness questions on ChatGPT every week, and that 70% of healthcare questions are now asked outside of clinical hours.
In ‘hospital deserts’ – areas located more than 30 minutes from a hospital – 4% of all ChatGPT queries are health related, he shared.
“Patients are taking care into their own hands,” Devraj emphasised.
Keep in the back of your mind how something like this could improve your productivity
AI: more than a large language model
Devraj emphasised that AI can do more than work as a large language model, which is the usage that many people might be the most familiar with.
Specialist tools for AI use in ophthalmology include RetFound and Google DeepMind’s MedGemma, he told attendees.
RetFound analyses ophthalmology images that it has specifically been trained on, Devraj explained.
Knowledge, agents and robotics are the phases that AI tools go through, he said.
Devraj added that the possibilities for such tools include addressing the ophthalmology backlog, improving patient care, and enhancing clinical productivity, whilst challenges include data generalisability, clinical integration, racial and ethical bias, workforce impact, legal uncertainty, and human factors such as acceptance and workflow.
Practitioners must question where a tool such as ChatGPT gets its data from, and how it came to its decision, Devraj said.
He also reminded attendees of the difficulties in regulating technology that is always changing.
Despite this, he emphasised his belief that AI tools could still become a practitioner’s ‘second eyes’ within the coming years.
Use AI to enhance your learning, and improve your confidence
MedGemma: the clinical assistant of the future?
Devraj’s session included a discussion where practitioners discussed case scenarios and saw fundus images analysed by AI.
Devraj allowed practitioners to interact with case scenarios through a website that he had previously created himself, using the open-access MedGemma model.
The MedGemma model has the functionality to create a patient summary, for example for a patient’s carer, and has 82% accuracy in passing ophthalmology exams, Devraj told attendees.
“While you’re doing this, keep in the back of your mind how something like this could improve your productivity,” he advised.
A model such as MedGemma could be acting as a clinical assistant to practitioners as early as 2028, Devraj believes.
In terms of whether the use of the MedGemma model could hinder learning and understanding, Devraj asked practitioners to ask themselves the question: “Are we using this as a tool, or as a replacement?”
Devraj hopes practitioners will think of it as a tool, he said.
“These sorts of models need to be validated,” Devraj emphasised, also noting that clinicians will often be training on AI tools in real time.
AI is often slow to be adopted in healthcare settings because there is a lack of existing regulation, he added.
He reminded attendees that it is important to always check “the patient sat in front of them – what is best for them?”
“You would not use it [AI] as a management tool,” Devraj emphasised: “you would use it as your assistant.”
“Consult it as a productivity tool, to extract what you need,” he advised.
He also noted that AI models are currently mainly being used in research, rather than in patient-facing settings, and that, where such models are used, every patient will need to provide consent.
Until there is regulation, AI models will not be widely used in patient settings, Devraj acknowledged, although he did note that “The College of Optometrists understand that there is an opportunity.”
Devraj also emphasised that no practitioner is expected to know everything there is to know about AI models right from the start.
“We learn at different speeds; we come in at different education and understanding levels,” he said.
“What I would do, at this stage, is use AI to enhance your learning, and improve your confidence.”
To experienced optometrists who might be concerned about embracing new tools, he said: “Somebody who is just going into the optometry workforce doesn’t have experience – and we know that experience is your best teacher.”
Comments (0)
You must be logged in to join the discussion. Log in