Using AI To Prevent Blindness
Healthcare is one of the most important fields Artificial Intelligence (AI) is going to transform. Diseases affecting the global population transversally are already on target. For instance, early diagnosis of Diabetic Retinopathy, leading cause of vision impairment globally, has been of primary interest for Google Health since circa 2016. In the AI team at Sngular ourselves, we have been working on the AI module for a prototype “Automatic Digital Ophthalmoscope VR”, a device capable of capturing an eye fundus image and providing a Diabetic Retinopathy diagnosis.
In this article, we share the insights gained from working on this use case and how AI can help prevent visual impairment and blindness caused by Diabetic Retinopathy.
What is Diabetic Retinopathy?
Diabetic Retinopathy is a leading cause of vision impairment and blindness in working-age adults. It affects about a third of people with diabetes (1 in 10 people have diabetes) and its damaging effects on vision can be prevented by early detection and treatment.
This condition arises when high glucose levels in the blood harm blood vessels in the retina, which is a layer of cells at the back of the eye that detects light. Affected blood vessels can swell and leak, causing blurry vision or stopping blood flow. Sometimes this promotes the excessive growth of new blood vessels, usually thinner and more prone to hemorrhages, feeding a vicious cycle of scar tissue development that can eventually cause retinal detachment.
Diabetes-related eye problems can be avoided by managing blood glucose levels, blood pressure and high fats. Medication, physical activity, and a healthy diet have proven effective in this regard. If it progresses to an advanced stage, treatment with laser and, if available, intraocular drug injections can prevent vision impairment and blindness. Examining patients in the early stages, while their vision remains strong, is crucial to benefit from the potential of less invasive treatment.
Examining patients in the early stages, while their vision remains strong, is crucial to benefit from the potential of less invasive treatment.
How is Diabetic Retinopathy currently diagnosed?
Diabetic Retinopathy is diagnosed through screening programmes using digital retinal photography. For example in the UK, all individuals with diabetes aged 12 years and over are invited for a diabetic eye screening appointment at least annually. Those considered to be at higher risk of progression of retinopathy can be invited for screening more regularly.
The clinician screening your eyes examines your retina and classifies it into one of the five disease stages shown in the figure below. They then develop an appropriate treatment plan.
How can AI help prevent blindness caused by Diabetic Retinopathy?
Vision problems can have major consequences in terms of use of health and social care resources and impact on economic productivity. Many countries provide eye checks for people with diabetes, but these are often insufficiently funded or organized systematically as a screening pathway. As a result, many people with diabetes are living with preventable vision impairment and blindness.
Many people with diabetes are living with preventable vision impairment and blindness.
When it comes to low-income and middle-income countries (LMICs), this challenge becomes more difficult because they have no or very limited screening services. The lack of trained personnel is a significant barrier to the provision of care for diabetic eye problems in these countries. For instance, in sub-Saharan Africa, around 16.3 million individuals are expected to have this ailment, but only 2.5 ophthalmologists exist per million population. This figure is much lower than the global average of 37.5.
The lack of trained personnel is a significant barrier to the provision of care for diabetic eye problems in LMICs countries.
AI can aid in halting the progression of Diabetic Retinopathy through automated diagnosis. The success of deep neural networks in image classification beyond the healthcare sector shows potential for their use in diagnosing eye fundus images. This can benefit regions without medical staff, such as LMICs, or aid ophthalmologists during diagnosis. AI could even help provide a first diagnosis to patients in primary care centres where no specialised staff is available. This could improve patient care and reduce diagnostic times.
There are already alternatives to the traditional Diabetic Retinopathy screenings involving AI in the market. Google’s ARDA is currently used to detect Diabetic Retinopathy in India and the European Union. In the United States, Eyenuk and iCare ILLUME already have FDA approval.
Another challenge to tackle is that nowadays medical check-ups create more data per patient than a doctor can study. AI can assist in processing this data and presenting an easily understandable summary to doctors. This can aid in anticipating medical incidents and even provide a medical pre-diagnosis through data analysis.
Nowadays, medical tests are generating more data a doctor can study. AI can assist in processing this data.
Predictive models can also aid in determining the ideal frequency for medical appointments personalized to each patient. The screening program for Diabetic Retinopathy in England is deemed a success. After implementing the program for fifty years, they announced it no longer caused the most cases of blindness in the working age population. They are now re-evaluating the frequency of medical appointments, which could be helped by an AI predictive model.
Conclusions
We have seen three ways in which AI can help prevent blindness, specifically, blindness caused by Diabetic Retinopathy:
- Providing an automatic diagnosis.
- Processing enormous amounts of medical data and presenting them to doctors in a manageable way.
- Designing a personalized screening program by optimizing medical appointment frequency on a per patient basis.
🔄 If you enjoyed the article, you can follow me on LinkedIN.🔄
