Skip to main content

The trial of an Artificial Intelligence retinal screening tool, developed in New Zealand, showed it did not miss any referable disease for diabetic patients at risk of eye disorders.

THEIA has been created by Toku Eyes and was tested for diabetic retinopathy. Co-founder Ehsan Vaghefi says the tool is being further developed to vastly improve early detection of eye disease such as macular degeneration and glaucoma.

Its use was trialed at Counties Manukau DHB in its central and satellite units, as well as a private optometrist in Palmerston North.

More than 900 patients, who are part of the diabetic screening programme, had their de-identified images screened as usual by retinal specialists, as well as by THEIA.

The study, conducted between March and May of this year, found the AI did not miss any patients with more than mild or vision threatening disease and the level of agreement with the clinicians was between 91-98 percent.

Vaghefi says people wait around six weeks to get their results from the current screening process, but with THEIA they would get them instantly. He says the tool could expand existing screening capacity by 50 percent and reduce labour costs by nearly a third, allowing more patients to access these services, including those in under-served areas.

Diabetic retinography is a leading cause of blindness in working age people, but there are screening backlogs nationwide.

Jacob Benefield, optometrist at Naylor Palmer Optometry in Palmerston North, says:

“Toku Eyes shows how AI can really help increase proactive retinal screening, maximising the time optometrists and ophthalmologists can spend with patients whose needs are more complex. Putting this tool into the hands of lab technicians and medical imagers will expand screening services, ensuring more people are seen sooner and improving their chances of early detection of diseases like diabetes,” 

Toku Eyes recently raised $3.6 million from investors to advance the commercialisation of THEIA and is applying for FDA approval in the US. Vaghefi says the AI platform is designed to handle lower quality images and anyone can be trained to operate the necessary machine, freeing up the time of specialists to focus on the 15-20 percent of people who need a consultation.

The use of AI also makes screening more accessible to people who struggle to attend hospital appointments. He would like to see a national diabetic screening programme developed with satellite units nationwide, supervised from the centre.

If you would like to provide feedback on this news story, please contact the editor Rebecca McBeth.

NEWS – editor Rebecca McBeth
Article originally on