The institute said their approach incorporated a specialised AI model that learns from past blood sugar trends and predicts future levels more accurately than existing methods.
“Unlike traditional forecasting models, which often struggle with long-term trends and require manual adjustments, this model processes glucose data automatically, identifying key patterns and making precise predictions,” the NIT researchers said, adding that basic neural networks often fail to recognise long-term glucose fluctuations.
NIT researchers argued that the current predictive AI models have a few drawbacks. “Many of these models work like a ‘black box,’ meaning their predictions are difficult to understand. This lack of transparency makes it hard for doctors and patients to fully trust them,” the institute said.
Diabetes is a major health challenge in India, with cases expected to reach 124.9 million by 2045. Effective diabetes management relies on regular glucose monitoring to prevent dangerous spikes – hyperglycemia, and drops – hypoglycemia, in blood sugar levels.
“Our core innovation lies in using multi-head attention layers within a neural basis expansion network, which allows the model to focus on the most relevant data points while ignoring unnecessary noise,” said Mirza Khalid Baig, assistant professor, biotechnology and medical engineering, and main author of the study. He explained that the new method results in better performance without the need for large amounts of training data or extensive computing power.
Discover the stories of your interest
Baig said in the long run, this AI-driven approach could be integrated into smart insulin pumps to automate insulin delivery. “It can be incorporated into mobile health apps for real-time glucose tracking, or used in clinical settings to support doctors in making personalised treatment plans,” he said.Currently, the researchers are planning on testing the developed technology through extensive clinical trials at hospitals, in collaboration with senior diabetologists.
