Image: © Ashley Gilbertson / Commonwealth Eye Health Consortium
Artificial intelligence (AI) has the potential to revolutionise many areas of healthcare. Alongside improving healthcare in high income countries, it promises rapid gains in low- and middle-income countries (LMICs), where AI tools can reduce the need for skilled clinical staff, especially in remote locations.
Despite this promise, implementation into health systems is so far limited. Ophthalmology is a speciality that could particularly benefit from AI, and within eye health diabetic retinopathy (DR) is a disease that shows great potential for AI tools.
DR is a common complication of diabetes and is a leading cause of blindness globally. Early screening and treatment can substantially reduce the risk of sight loss in people with the disease. By submitting retinal photographs from people with diabetes to an AI tool, grading of DR, where the severity of the disease is assessed, can be automated. This can reduce the burden on specialist eye care staff and streamline referrals to other services by providing a diagnosis at the point of screening.
To ascertain the progress of this new technology in low-income settings, and identify areas for future research, a new study by ICEH looked at how AI has been used for DR in LMICs.
The review article included 81 papers. Most of these focused on the performance of AI tools for interpreting retinal photos and providing a grade. Others had different specific outcomes such as grading other retinal conditions at the same time, or looking at macular oedema, a swelling of the retina that can be caused by different diseases, including diabetes.
The authors found that the majority of papers reported how accurate the AI tools were for detecting and grading retinal images for DR, with promising results. However, there was limited evidence for the implementation of these tools into clinics and whether they led to improvements in eye health services or patients’ vision.
Only three of the studies looked at whether and how an AI tool could be successfully implemented into clinical services. A number of challenges were noted including issues such as electricity and internet access. One of these studies from Rwanda showed that when AI-assisted DR screening was combined with a referral decision on the day, the proportion of people attending the specialist eye clinic was increased, indicating the value of a point-of-screening result.
Nearly two thirds of studies came from India, China and Thailand, all middle-income countries, highlighting a lack of research in low-income settings. If this trend continues it could lead a widening of the digital health divide.
None of the studies looked at different AI tools on the same dataset, making it difficult to make a meaningful comparison between different systems.
Furthermore, other crucial factors such as information on regulatory approvals and the cost of using the AI tools were not easily available. Other considerations include legal and ethical issues, which are important for data storage and accountability for errors. Improved screening also needs to be coupled with improved access to treatment for DR, the authors note.
Overall, the study highlights the significant potential for AI to improve diabetic eye care in LMICs. However, there is a need for more studies to evaluate the impact of AI on eye health programmes and visual outcomes. If AI can be shown to improve outcomes in low-income countries whilst being integrated appropriately and cost-effectively into existing health systems, it will be a valuable technology to improve and strengthen eye care.
Cleland CR, Rwiza J, Evans JR et al. Artificial intelligence for diabetic retinopathy in low-income and middle-income countries: a scoping review. BMJ Open Diabetes Research & Care. August 2023. https://doi.org/10.1136/bmjdrc-2023-003424