AI-powered tool GluFormer offers early warning and tailored care for diabetes

The model was trained on a dataset of over 10,000 non-diabetic participants, ensuring it captures diverse patterns and generalizes well to different populations.

By Pesach Benson, TPS

A groundbreaking AI model developed in Israel is offering hope to millions of people managing diabetes or monitoring blood sugar levels by predicting glucose responses to foods, identifying early diabetes risks, and enabling tailored treatments.

Developed by researchers from the Weizmann Institute of Science, Tel Aviv-based startup Pheno.AI, and the US-based NVIDIA, the tool called GluFormer uses artificial intelligence to predict glucose levels and other health metrics based on past monitoring data, providing personalized insights into health management.

Diabetes currently affects approximately 10% of adults worldwide, with numbers projected to double by 2050 to over 1.3 billion people.

The disease is a leading cause of death globally and can result in severe complications such as kidney damage, vision loss, and heart problems. The economic burden of diabetes is expected to reach $2.5 trillion globally by 2030.

GluFormer builds on the data from continuous glucose monitors (CGMs), wearable devices that track blood sugar levels in real-time.

CGM data has already been instrumental in diagnosing prediabetes and diabetes earlier, but GluFormer elevates this potential by forecasting health outcomes up to four years in advance, helping clinicians and patients detect irregularities and plan preventative care strategies.

By incorporating dietary intake data, the AI tool can also predict how specific foods or diet changes will impact an individual’s glucose levels. This feature opens doors for precision nutrition, enabling users to make informed dietary choices tailored to their unique metabolic responses.

“Medical data, especially from CGMs, can be viewed as sequences of diagnostic tests that map biological processes over time,” explained Gal Chechik, senior director of AI research at NVIDIA.

“The transformer architecture, originally developed for analyzing long text sequences, can also predict the trajectory of medical diagnostics, revealing valuable insights into a patient’s health.”

The novel technology behind GluFormer lies in its use of transformer model architecture, a cutting-edge neural network design originally developed for processing sequential data, such as text, in models like OpenAI’s GPT.

This marks a significant leap in its application to healthcare, specifically for continuous glucose monitoring (CGM) and broader medical diagnostics.

GluFormer was trained using CGM data collected every 15 minutes over 14 days. This granular, real-time data enhances the model’s ability to detect subtle trends and irregularities, making it more effective than conventional diagnostic tools.

By incorporating dietary data into its analysis, the model can predict individual responses to specific foods, enabling tailored health advice.

The model was trained on a dataset of over 10,000 non-diabetic participants, ensuring it captures diverse patterns and generalizes well to different populations.

The researchers also validated the model on 15 other datasets, demonstrating its ability to predict outcomes for people with various conditions, such as prediabetes, type 1 and type 2 diabetes, obesity and gestational diabetes.

In addition to glucose levels, GluFormer can forecast various other medical metrics. These include the apnea-hypopnea index, which is a measure of sleep apnea and is associated with type 2 diabetes.

It can also forecast systolic blood pressure, which is linked to diabetes risk, and visceral adipose tissue (a marker for fat around organs).

“The convergence of generative AI technology and large-scale health data collection enabled us to extract valuable medical insights. It’s a significant step forward in leveraging AI for healthcare,” said lead researcher Guy Lutsker, a Nvidia researcher and Ph.D. student at the Weizmann Institute.

Predicting individual glucose responses to specific foods will allow patients to make informed dietary choices to prevent spikes or crashes in blood sugar levels.

For those at high risk of developing diabetes, the model can help identify early warning signs, enabling timely lifestyle adjustments or medical interventions.

GluFormer also opens the door to tailored treatment plans and allow doctors to track how medical interventions impact a patient’s glucose levels over time.