Experts believe this capability could make a significant difference in countries like India, where access to regular blood testing may be limited in rural or underserved areas
In a development that could transform diabetes screening and early diagnosis, Indian and US researchers have created an artificial intelligence (AI)-based technique that can detect diabetes using a simple eye scan, eliminating the need for traditional blood tests.
The method relies on analysing high-resolution photographs of the retina—the light-sensitive layer at the back of the eye—to determine whether a person has elevated blood sugar levels. By examining subtle changes in retinal blood vessels, the AI system can distinguish between people with and without diabetes, offering a non-invasive and potentially scalable screening tool.
The study, published in the journal Diabetes Technology and Therapeutics, demonstrates how AI can identify microscopic warning signs in the eye’s vascular structure that are not visible to the human eye. These minute variations in the shape and curvature of retinal blood vessels, known as vessel tortuosity, can reveal metabolic changes linked to diabetes long before clinical symptoms appear.
“India has over 100 million people living with diabetes, and a significant number remain undiagnosed,” said Dr. V. Mohan, Chennai-based diabetologist and Padma Shri awardee, who was part of the research team. “If AI tools using simple retinal photographs can support early diagnosis, this approach could one day be deployed in real time to screen large populations.”
The research was led by a multidisciplinary team from India and the United States, including scientists from Emory University in the US and Yenepoya (Deemed to be) University in Karnataka. According to Dr. Sudeshna Sil Kar of Emory University, the AI was trained to identify specific patterns and shapes in retinal veins by comparing images from people with and without diabetes.
To develop and validate the technique, the researchers analysed 273 retinal images obtained from 139 participants. Using machine vision-based methods, they extracted 226 quantitative features related to vessel tortuosity, studying arteries and veins separately. These features were then fed into an AI model trained to recognise patterns associated with abnormal blood sugar levels.
The results were striking. In the test group, the AI system correctly identified individuals with diabetes with a sensitivity of 95 per cent. Importantly, it was also able to detect prediabetes—a critical early stage where lifestyle changes such as improved diet and increased physical activity can delay or even prevent the onset of full-blown diabetes.
Experts believe this capability could make a significant difference in countries like India, where access to regular blood testing may be limited in rural or underserved areas. Unlike conventional diagnostic methods, the AI-based approach does not require fasting, laboratory infrastructure or finger-prick blood samples. A quick photograph of the back of the eye is sufficient, making the process faster, more comfortable and potentially more cost-effective.
“This technology could serve as an effective, non-invasive screening tool,” the researchers noted, adding that it may be especially useful in community health settings, primary care clinics and large-scale screening programmes.
However, the experts cautioned that further work is needed before the technique can be adopted widely. The current findings must be validated across larger and more diverse populations to ensure accuracy, reliability and generalisability. Factors such as age, ethnicity, co-existing eye conditions and image quality will also need careful evaluation.
If future studies confirm these results, AI-powered retinal imaging could become a powerful addition to public health strategies aimed at tackling the growing global diabetes burden. By enabling earlier detection and intervention, the technology holds promise not only for improving individual outcomes but also for reducing long-term healthcare costs associated with diabetes-related complications.
