Artificial intelligence – computer programs that are trained to learn and complete tasks that would usually require human intelligence – is changing the landscape of medicine, and eye health is at the forefront.

CERA Principal Investigator Professor Mingguang He and his team have developed a cutting-edge artificial intelligence screening tool that can detect signs of glaucoma, macular degeneration, diabetic retinopathy and cataracts.

After taking a photo of the back of a patient’s eye, the AI system scans for signs of disease, and prints out a report identifying if the patient should be referred to a specialist for further assessment and treatment.

This would allow preliminary eye health screening to be done in a wider range of health settings, such as in GP clinics or by health workers in remote communities.

“Vision impairment and blindness are major public health problems in Australia, with up to 50 per cent of major eye diseases remaining undiagnosed,” says Professor He.

“Artificial intelligence has the potential to close the gap in eye care services, considerably increase early diagnosis of the four most common blinding eye diseases, and reduce the burden of vision loss in the Australian communities that need it the most.”

The tool has been found to be highly accurate and is currently being trialled in real-world settings.

The next step – predicting disease progression

Now, with the support of a National Health and Medical Research Council (NHMRC) Investigator Grant, Professor He is working to evolve this deep learning technology to develop a clinical decision system that is able to predict disease outcomes and prognosis.

“The potential of an AI application such as this is yet to be fully realised,” says Professor He.

“In the next five years, we will develop a system that will help clinicians decide the best treatment option, based on real-world clinical data.”

The current system can make a binary classification of specific eye diseases – for example, it might give a classification of no glaucoma, probable glaucoma or certain glaucoma. However, severity is based on a single image.

The new algorithm Professor He strives to create could provide a more detailed classification – for example, showing what type of glaucoma is present, and how likely it is to progress.

“This development would require access to data from additional imaging methods such as visual field, ocular coherence tomography (OCT), as well as real-world clinical data in text format, and to use this data to train a new AI algorithm to build a clinical decision system,” Professor He says.

“This system will use one particular disease, for example glaucoma, and will further expand to other diseases using the same framework and strategies when validation is proven.”