A team of Chinese and Singaporean scientists has developed an artificial intelligence (AI) tool for primary diabetes care, the world’s first multimodal large language model (LLM) designed specifically for controlling this common chronic condition.
Diabetes was expected to affect over 500 million people worldwide in 2021, with the majority of them living in low- and middle-income nations. These areas frequently have a shortage of educated primary care physicians and limited access to effective screening for diabetic retinopathy, a dangerous eye disorder connected with diabetes.
Researchers at Tsinghua University, Shanghai Jiao Tong University, and the National University of Singapore created a GPT-4-like system that may provide primary care physicians with tailored diabetes treatment assistance.
According to a recent study published in the journal Nature Medicine, the image-plus-language platform, known as DeepDR-LLM, is intended to leverage the power of LLM and deep learning to provide a comprehensive solution for medical image diagnostics and the delivery of tailored treatment recommendations.
The team utilized an open-source LLM that analyzed 371,763 real-world management tips from 267,730 participants. They then verified the picture module with 21 datasets containing standard or portable retinal images from seven countries: China, Singapore, India, Thailand, Britain, Algeria, and Uzbekistan.
According to the study, in a retrospective review, the system performed comparably to primary care physicians in English and outperformed them in Chinese.
Furthermore, in the task of diagnosing diabetic retinopathy, primary care providers’ average accuracy was 81.0 percent unaided and climbed to 92.3 percent when supported by the system.
The researchers urge for the integration of this approach into primary care diabetes protocols, as it has the potential to significantly improve the efficiency of both diagnosis and treatment, resulting in better health outcomes for diabetic patients.
