New Delhi, Jan 28 (IANS) A team of Indian and US researchers has developed an artificial intelligence (AI)-based technique to detect diabetes without the traditional blood tests.
The technique can detect whether a person has high blood sugar by taking a high-resolution photo of the retina (back of the eye).
The study, published in the Diabetes Technology and Therapeutics journal, showed that AI can spot tiny warning signs in the eye’s blood vessels that are invisible to the human eye, which can differentiate people with and without diabetes without a finger-prick blood test.
“India has over 100 million people with diabetes, and very often, many do not even know they have it. If the use of AI tools with simple retinal photos can help early diagnosis of diabetes, it can be used in real-time in the future to screen for diabetes,” said Dr. V. Mohan, a Chennai-based diabetologist and a Padma Shri awardee, who was part of the study.
Dr. Sudeshna Sil Kar, from Emory University, US, shared that the researchers trained the AI to look at specific shapes and patterns in the veins using retinal photos of people without and with diabetes.
The team, including those from Yenepoya (deemed to be) University in Karnataka, analysed 273 retinal images from 139 participants. They extracted 226 quantitative vessel tortuosity features separately for arteries and veins using machine vision-based approaches.
The AI technique was precise in correctly identifying diabetes using retinal photographs in the test group with 95 per cent sensitivity. The system could even spot ‘prediabetes’, which is the stage where lifestyle modification can help prevent diabetes.
It could prove to be an effective non-invasive mode of detecting diabetes early, as it does not require expensive laboratory equipment. The patients also do not need to fast or undergo a blood test. The method just requires a quick photo of the back of the eye, said the researchers.
The expert, however, stressed the need for validating the research findings in a larger population.
–IANS
rvt/
