Skip Ribbon Commands
Skip to main content

Research: Using AI to screen for visually significant cataracts in the primary health care

Age-related cataracts are the leading cause of visual impairment globally, accounting for 94 million adults aged ≥50 years who experienced low vision or blindness in 2020. Although a cataract is easily treatable, a significant number of patients with a visually significant cataract (that is, a cataract with severe visual loss) remain undiagnosed in communities, especially in rural areas, due to the limited availability of, or accessibility to, cataract screening. 

In a latest study by Prof Cheng Ching-Yu's group that was published in Nature Aging, they have reported the development and validation of a retinal photograph-based deep learning algorithm that can automatically detect, visually significant cataracts, using more than 25,000 images from population-based studies.

When they compared the AI algorithm with 4 ophthalmologists' evaluations, their AI system performed consistently better, if not comparable (sensitivity of 93% Vs 52-97% by ophthalmologists and specificity of 99% vs 91-98% by ophthalmologists). There seems to be a huge potential of a retinal photograph-based screening tool for visually significant cataracts among older adults, that could lead to more appropriate referrals to tertiary eye centres.

Image and paper source:                             

For more information, watch this video as Prof Cheng Ching-Yu discusses on his research work:
Internet access required


Interview responses from Prof Cheng Ching-Yu and Dr Tham Yih Chung:

Hi. Can you explain more about the objective of your study that can detect visually significant cataract using retinal photograph-based deep learning?

Visually significant cataracts are those associated significant vision loss so much so that it affects their quality of life and ability to carry out their daily functions. It is important to identify these cases and recommend them for care as cataract surgery may help restore vision and help improve quality of life. The aim of this study is to see if we can detect these cataracts with just a retina photograph.

How can this deep-learning system improve upon our existing diagnostic tools for cataract?

We hope to use this AI technology to 'detect' significant cataract at the community / primary care level (e.g. at polyclinics), so that these people can be referred onto tertiary care. There is already established use of retinal photography for eye screening at the primary care level (especially for diabetic retina disease) and the same process can be extended to screening for visually significant cataract.  

How does this deep learning system compare with ophthalmologist's evaluation in a face-face setting?

In our study, we compared the performance of our AI with 4 Ophthalmologists with specialist clinical experience ranging from 2 years to 8 years, and we were able to show that the AI preformed as well as, if not better than, our ophthalmologists!

Is it ready for real-world application yet?

Our next step is to test out this AI model in a real-world community setting and evaluate its real-world effectiveness and deployment feasibility. If successful, this will mean that in the future, if the primary physician (e.g. General practitioner) is uncertain if his / her patient has a visually significant cataract, he / she could send the patient to be screened by retinal photography and get a quick answer. This will improve the quality of referrals and decrease unnecessary referrals (e.g. those that may just need a change in their glasses).

Back to EyeSight main page.