Managing diabetes with digital tech innovation
Local doctors have teamed up with experts from home and abroad to develop an innovative system of combining language and image artificial intelligence model for the screening, diagnosis and management of diabetes – effectively using digital technology to tackle a major chronic disease.
Diabetes is one of the fast-growing chronic diseases in the world. It can cause various consequences like blindness, kidney failure, stroke and heart attack. There are over 500 million people with diabetes in the world and 80 percent of them live in low- and middle-income countries.
China is the home of about 140 million diabetics, topping the world.
![Managing diabetes with digital tech innovation](https://obj.shine.cn/files/2024/07/22/81251ee6-6e8a-4619-b5ef-2d343127109a_0.png)
Local doctors have teamed up with experts from home and abroad to develop an innovative project of combining language and image AI model for diabetes screening, diagnosis and management. The study was published in the world-leading journal Nature Medicine.
Primary diabetes care and diabetic retinopathy (DR) screening persist as major public health challenges due to a shortage of trained primary care physicians, particularly in low-resource settings, posing a big challenge to the public health system, both in China and the world.
DR is a major diabetes-led eye complication leading to vision injury and even blindness.
To bridge the gaps, experts from Shanghai No. 6 People's Hospital led a project to develop an integrated image-language system DeepDR-LLM, combining a large language model (LLM module) and image-based deep learning (DeepDR-Transformer), to provide individualized diabetes management recommendations to primary physicians at the grassroots.
In a retrospective evaluation, the system showed great effects in assisting primary physicians. For identifying referable DR, the average primary physicians' accuracy was 81 percent unassisted and 92.3 percent with the assistance of the system.
Furthermore, experts performed a single-center real-world prospective study, deploying DeepDR-LLM. They compared diabetes management adherence of patients under unassisted physicians with those under the system.
Patients with newly diagnosed diabetes with the system showed better self-management behaviors throughout follow-up. For patients with referral DR, those in the management of the system were more likely to adhere to DR referrals. Additionally, DeepDR-LLM deployment improved the quality and empathy level of management recommendations.
"Given its multifaceted performance, we are confident that DeepDR-LLM holds promise as a digital solution for enhancing primary diabetes care and DR screening," said Dr Jia Weiping from Shanghai No. 6 People's Hospital, a leading expert on the project.
The research has received wide recognition and was published by the world-leading journal Nature Medicine. Its DeepDR-Transformer module has been developed and validated in 14 datasets across five countries (China, Singapore, India, Thailand and the United Kingdom) with standard fundus images, and seven datasets across three countries (China, Algeria and Uzbekistan) with portable fundus images, experts said.
![](https://obj.shine.cn/website/assets/logo-s.png)