AI Revolutionizes Diabetes Management: Predicting the Future with GluFormer

Is artificial intelligence about to transform the diagnosis of diabetes? Israeli researchers have announced the development of a groundbreaking AI model named GluFormer, capable of accurately predicting the blood sugar levels of patients up to four years ahead. Trained on data from thousands of participants, the model can identify patterns and anticipate how the body will react to different foods.

The Future of Diabetes Management

The remarkable model, known as 'GluFormer', was created in collaboration with three leading institutions: the Weizmann Institute of Science, led by Prof. Eran Segal from the Department of Computer Science; NVIDIA Israel: the research group at NVIDIA, led by Prof. Gal Chechik, which provided the technological infrastructure and support; and Pheno.AI - a healthcare AI startup that contributed its expertise and experience to develop the model.

The model, based on an architecture similar to that of large language models like ChatGPT, "learns" patients' sugar profiles from past data and translates this information into accurate forecasts. "Participants wore a continuous glucose monitor for two weeks, during which the technology analyzed and predicted future measurements," explained Prof. Gal Chechik, head of AI research at NVIDIA Israel. The model was trained on over 10 million glucose level data points collected every 15 minutes using a wearable monitoring device. The idea is that now data analysis can happen similarly to language-based AI models, except in this model, the technology analyzes previous glucose readings and attempts to predict future ones.

"When people think about AI, they immediately think of language models like ChatGPT, for example," added Chechik, "and indeed the working method is similar, but in this case, we taught a similar model to handle glucose measurements. It learns how sugar profiles develop in a person over time."

Researchers noted that GluFormer has numerous predictive abilities, including forecasting blood sugar levels, allowing it to predict blood sugar up to four years into the future. It also offers personalized adaptation by analyzing how the body reacts to different foods, enabling tailored nutrition. Additionally, it can predict other medical metrics, such as fat tissue, systolic blood pressure, and sleep apnea index. The model also aids in early diagnosis of pre-diabetes and diabetes, thus holding the potential to assist hundreds of millions of people worldwide dealing with diabetes.

Researchers claim they believe future developments of GluFormer will enable the prediction of additional medical metrics and even contribute to new drug development. "Just as we taught the model on glucose measurement results, technology can be developed to teach the language of proteins and different molecules, helping us, for example, to create new medicines," clarified Prof. Chechik.