In the UK, the National Health Service (NHS) invites diabetic patients for eye screening every one or two years. The US guidelines recommend annual screenings for adults with type 2 diabetes starting at diagnosis. Despite these guidelines, many patients miss these essential screenings due to barriers such as cost, communication, and accessibility. Roomasa Channa, a retina specialist, emphasizes that regular screening can prevent vision loss by detecting diabetic retinopathy early.
Screening involves photographing the fundus, the eye's interior rear wall. Traditionally, this requires manual interpretation of fundus images, which is repetitive and time-consuming. However, AI technology presents a promising solution by automating the detection of diabetic retinopathy. AI can be trained to recognize the disease's stages, potentially determining whether a referral to an eye specialist is necessary or assisting human image graders.
One such AI system is developed by Retmarker, a health technology company in Portugal. Their system identifies potentially problematic fundus images for further human expert evaluation. Retmarker’s CEO, João Diogo Ramos, notes that while the AI tool supports human decision-making, resistance to change limits its widespread adoption. Independent studies indicate that systems like Retmarker Screening and Eyenuk’s EyeArt achieve acceptable sensitivity and specificity rates. Sensitivity measures the test’s ability to detect disease, while specificity measures its ability to confirm the absence of disease. High sensitivity can lead to false positives, which may cause unnecessary anxiety and specialist visits.
Google Health’s AI system for detecting diabetic retinopathy faced challenges during trials in Thailand due to variable image quality from different lighting conditions and camera operator expertise. This highlighted the need for high-quality data and diverse user feedback in developing effective AI models. Google’s system has been licensed for use in Thailand and India, with the Thai Ministry of Public Health assessing its cost-effectiveness.
Cost is a crucial factor in the adoption of AI screening tools. Retmarker’s service costs around €5 per screening, though prices vary based on volume and location. In the US, higher medical billing codes increase costs. In Singapore, researchers compared the costs of three diabetic retinopathy screening models: human assessment, full AI automation, and a hybrid model combining both. The hybrid model, which uses AI for initial filtering followed by human assessment, proved most cost-effective and has been integrated into Singapore’s national IT platform for healthcare.
Bilal Mateen, PATH’s chief AI officer, argues that while AI tools have shown cost-effectiveness in affluent countries, their impact in lower-income regions remains uncertain. He stresses the importance of developing AI solutions that are accessible to all, not just privileged populations. Dr. Channa concurs, highlighting the need to bridge health equity gaps within countries like the US. She advocates for expanding AI screening tools to regions with limited eye care access, ensuring they supplement rather than replace comprehensive eye examinations.
Despite its potential, AI struggles with detecting more complex eye conditions like myopia and glaucoma. Nevertheless, Dr. Channa finds the technology’s potential to streamline diabetic retinopathy screening exciting. She hopes AI will enable timely screenings for all diabetic patients, ultimately reducing the burden of diabetes-related vision loss.
In Yorkshire, Quinn reflects on the potential of AI in earlier detection of diabetic retinopathy, wishing such technology had been available when he first developed the condition. His experience underscores the transformative impact that early detection and treatment can have on preserving vision and improving the quality of life for individuals with diabetes.
