Home » Health and social care » Artificial intelligence is changing the way we diagnose eye disease
Health and social care

Artificial intelligence is changing the way we diagnose eye disease

AI is transforming the way healthcare practitioners and doctors detect disease, enabling quicker diagnosis and therefore more effective treatment plans for patients. This is particularly true for eye disease.


Artificial intelligence is highly topical in the world of healthcare and for all the right reasons.

For ophthalmologists (the eye doctors that perform surgery on the eyes and manage patients with eye disease) artificial intelligence tools are starting to help interpret the millions of photographic images that are produced in today’s eye centres in the UK.

No human could look at a retinal photograph and accurately generate such information.

Eye patients regularly have digital images taken of their eyes to diagnose and assess possible disease such as macular degeneration. These images include photographs of the retina, the tissue that senses light. Cross-sectional images of the retina are also regularly obtained. These advanced optical coherent tomogram (OCT) scans show all the different cellular layers the retina contains and can have over a million data points in one single image.

Retinal photographs are an important part of the national diabetic retinopathy screening programme in the UK. With diabetes on the increase, many people do not realise that this condition can lead to sight loss if it’s not treated.

Diabetes can affect eyesight through your blood vessels

Diabetes affects small blood vessels, damaging the retina. When the blood vessels in the central area of the retina (the macula) are affected, it’s known as diabetic maculopathy. Screening is a way of detecting the condition early before any noticeable changes affect vision. Diabetic retinopathy doesn’t usually cause any noticeable symptoms in the early stages.

The convergence of increasing computing power and the ability to access and quickly diagnose the millions of scans by a limited number of eye doctors and technicians , would be almost impossible. This is why computers, and the way we use their learning abilities, can make such a difference to the number of patients that can be diagnosed early and eventually treated to save their sight.

AI can detect your age, gender, blood pressure and smoking habits from retinal photographs

An early example of the power of this approach includes a study of retinal photographs where, after a period of training on several hundreds of thousands of retinal photographs, computers were able to analyse photographs of an eye and very accurately predict the age of the patient, their gender, their blood pressure, whether they smoked and what their blood sugar measurement was likely to be. No human could look at a retinal photograph and accurately generate such information.

A collaborative study between Moorfields Eye Hospital and Google DeepMind demonstrates how a computer algorithm could analyse OCT images and make accurate referral suggestions, which are comparable to the standards of a clinical experts; therefore deciding whether that patient has a sight threatening problem that requires urgent attention or not.

AI can help experts diagnose more patients more efficiently

What this means for the future is that computer algorithms generated by artificial intelligence will help manage patients more speedily and as accurately as clinical experts. This will help us manage the ever-increasing number of patients who have serious eye problems such as macular degeneration or diabetic retinopathy.

The power of artificial intelligence will also help us better understand the fundamental cause of diseases too, and institutions are investing in the future.  The Wellcome Trust has recently funded a project to use artificial intelligence to find the earliest changes in the visual system that cause macular degeneration. The exciting expectation is that the research, which I am leading, will ultimately develop new treatments to save sight.

Next article