Skip Ribbon Commands
Skip to main content

On a mission to save sight using artificial intelligence

Dr Daniel Ting, Associate Consultant, Singapore National Eye Centre, hopes that with the advent of artificial intelligence (AI) and its applications in healthcare, more can be done to save the sight of those who may be afflicted with serious eye diseases. 

Tell us more about the importance of eye screening.
For diabetes patients, regular eye screening is essential as symptoms of diabetic retinopathy usually do not show up until the disease is too advanced to treat and patients lose their sight. To counter this, every year, more than 100,000 patients under go eye scans where photos are taken of their retinas (back of the eye) under the Singapore Integrated Diabetic Retinopathy Screening Programme (SiDRP). Optometrists and eye care professionals then check each photo for signs of diabetic retinopathy, glaucoma, age-related macular degeneration (AMD), and other retinal conditions. A report is sent back to the hospital, polyclinic or GP within one day.

Usually, about 60-70% of the patients have no signs of eye disease. Where scans indicate an eye condition may be present, they are re-checked to confirm that these patients need additional evaluation and treatment. Thereafter, prompt referrals are made for patients who require specialist care. 

Your research into using AI for eye screening sounds very exciting. How did it start and how do you think it will transform the current way of doing eye screening?
We first started exploring the use of AI for eye screening in collaboration with the National University of Singapore's (NUS) School of Computing. The NUS team built the AI’s ‘brain’ while we provided the ‘dictionaries’– about 275,000 retinal scans from healthy and diseased eyes – so that the system could identify the difference between ‘normal’ and ‘abnormal’ scans. 

To test the system, we used it to scan another 200,000 multi-ethnic retinal samples and validated the results against those manually graded by retinal specialists, ophthalmologists, optometrists and eye care professionals. We were very excited when we found that our AI system achieved more than 90% accuracy in most datasets and subsequently published our findings in the Journal of American Medical Association  in December 2017. We continue to feed data from on-going retinal screenings into the system to sharpen the performance of our AI algorithm.

We are harnessing AI capabilities to conduct the first round of checks for all retinal images taken to identify signs of diabetic retinopathy, glaucoma and AMD in retinal scans. This will make diagnosis cheaper and faster than the current way of screening which relies on human skill. 

What other plans are in the pipeline for the use of this AI system?
We aim to use AI to conduct the first round of checks for all retinal images taken under the national screening programme. This will enable patients who have no abnormalities to receive their results instantly. More importantly, it will allow our optometrists and eye care professionals to focus on confirming the accuracy of scans that are found to show abnormalities and, of course, to treat patients who need specialist care. Ultimately, we hope that we can harness the capabilities of AI for the entire screening process, without any compromise in accuracy and quality as we continue to keep a close watch on the clinical data output.  This will allow us to more efficiently manage the increasing demand for retinal screening, especially as we are confronted by an ageing population. 

Interestingly, we noticed that the AI system is very sensitive and can pick up subtle signs of eye disease that may be invisible to the naked eye. This means we may possibly be able to detect minute changes in the eye even before the progress of disease. At the moment, we are exploring ways to stratify patients into different risk profiles so that we can make medicine more personalised to diabetic patients. 

This may eventually profoundly impact those afflicted with diabetes worldwide and prevent vision loss associated with this chronic condition.