The eyes are the window to… cardiovascular disease? 

The father of Toku Eyes co-founder Ehsan Vaghefi lost his eyesight when he was four years old in Iran as a result of congenital glaucoma.

If the disease had been diagnosed and treated early, his vision could have been saved but his family simply didn’t have the money for the costly treatment.  

“That shaped the story of my life,” says Vaghefi. 

After migrations to four different countries, “one thing led to another”, and Vaghefi landed in New Zealand, taking up a position in the optometry and vision science department at the University of Auckland.

While working on imaging technology for his PhD, Vaghefi experienced a “lightbulb moment,” discovering that the transparency of the eye makes it easy to photograph blood vessels, in turn revealing early signs of cardiovascular problems or conditions like diabetes.

He’d also met David Squirrell, a University of Auckland academic and lead ophthalmologist for the national Diabetic Eye Screening programme who said there were too many patients in his clinics for the number of staff.

The pair realised their combined expertise and artificial intelligence could make a difference and in 2019 they co-founded Toku Eyes, a medtech company developing non-invasive, AI-powered diagnostic and screening tools using retinal imaging with a goal of making ocular care more accessible to those on lower incomes.

“Back then, AI was very new, at least in ophthalmology,” says Vaghefi, who is CEO. 

Investors in the Auckland University spin-out include Uniservices, Icehouse Ventures, Artemis Capital and United States non-profit Kera Link International – a social venture that invests in technology to address blindness in developing countries. Toku has also received about $4 million in grants from various organisations, including the Health Research Council, MBIE and Callaghan Innovation.

In April 2023, Toku, which is now a US-registered company, closed a US$8m Series A preferred financing round. The financing was co-led by National Vision, one of the largest optical retailers in the United States, and Topcon Healthcare, a leading provider of medical devices and software solutions for the global eye care community. 

Though Vaghefi says each company initially wanted to buy the startup, he didn’t want to sell. 

“Thankfully, they believed in me, and hopefully, I’ve done enough to prove that,” he says. 

The capital injection helped move the company's headquarters to San Diego but the 20-strong Toku team is spread worldwide. Keeping R&D in New Zealand, the startup has a second North American office in Atlanta and one in London, United Kingdom (UK). It will soon be opening an office in Dubai. 

Having recently cleared regulatory hurdles in the US under a fast track clearance, Vaghefi says Toku Eyes is “just getting to market” and will soon be generating revenue. 

Vaghefi sits down with Caffeine to discuss the ins and outs of AI as a medical tool. 

How are you using AI in your business? 

AI comes in many flavours. In our case, AI is your doctor, making a decision about your health and generating a referral for you. 

To do so, we have put together one of the largest databases of its kind from all around the world – I believe we have the images of more than two million people’s eyes in there. With that, we know everything there is to know about them: their blood pressure, the medications they’re taking, if they’ve ever had a stroke, and so on. 

The outcome of our AI is if you are at risk, it will tell you to go and see a cardiologist. 

‍

How do you foster AI literacy within your organisation?

Everyone comes in knowing something about AI, but because our technology is so niche, that doesn’t mean they know what we need them to know. There’s a lot of learning, and we pride ourselves on being able to teach people something no one else can. 

We have four teams within our small group, and each of them knows something different. Not everybody in an AI company needs to know the same thing about AI. 

There are the core engineers – these are the people who have actual PhDs in AI. They know how to code and how to create pure AI. 

Then there are the clinical AI people, who ensure the AI doesn’t break the workflow – if the AI takes five minutes to generate results, it will never work in a clinical environment because all you have is thirty seconds. 

Then, we have the actual product team, who help capture more information from the patient to enhance the results. 

The final piece is regulatory. As our medical device works with the Food and Drug Administration (FDA), every time we change something in the AI, we have to report it. Because we’re so cutting-edge, we often create things as we go. We wrote the guidelines for the FDA because it’s so new. 

‍‍

What has been your biggest challenge with AI?

The science is straightforward; creating an AI that does a certain thing is straightforward; the regularity has been tough but a good experience because we taught them what we learned ourselves. 

The biggest challenge has been making an AI that enhances someone’s clinical experience and doesn’t break things. 

When we started in California, we were talking to this multibillion-dollar business. I met with the CEO, and he said, ‘I love your technology; we will roll it out in our clinics. But if the AI adds one minute to my patient time, I will kill it because it means I see one less person by the end of the day, and I cannot tolerate that.’ 

It was a simple statement but a considerable challenge. We went back to the drawing board and asked ourselves how we could achieve the performance and results they needed with fewer resources. 

How do you manage privacy and legal concerns? 

We deal with privacy exactly as the law demands it. The law is not challenging; you just have to understand the regulations. 

We are a medical device company; our AI is a medical device. In different jurisdictions, our AI is regulated by the governing bodies. We have a CE Mark, UKCA Mark and we just got our FDA breakthrough designation. 

Some people seem stuck in 2018-2019 when all these AIs were learning from everything, and suddenly, there were breaches. Today, there are very good laws; you just have to abide by them. I’m not saying it’s easy, but it’s straightforward.

‍

Does AI legislation help or hinder you? 

It helps, absolutely. When people look at a challenge, they look at it as something they should overcome. I look at it as a moat around my business. If I clear it, the next person has to clear it too, so the more regulated things are, the better for me. 

In New Zealand, I’m trying to set up a system. In Australia and New Zealand the regulatory system tries to reinvent the wheel but the laws are there; there is a European law and an American law. Just copy and paste it with modifications. 

‍

Are there any AI trends that you are closely monitoring? 

The large language models. 

My AI, and others like ours, are in decision-making. To me, that is done and dusted; enough research and trial has been done, and now, we’re at commercialisation. 

The large language models will change the relationship between the clinician and the patient. Look at an average GP appointment; you probably have 20 minutes. Half the time is spent telling the GP what’s wrong, five minutes are spent on different measurements, and the last five minutes are spent on decision-making and prescription writing. 

The 20-minute appointment could be reduced to 10 if you talked to a machine, which can reduce and transfer the information into exactly what the doctor needs. Then, the clinician can make a decision in two or three minutes, and the large language model can generate a prescription.

‍