Can machine learning help manage type 1 diabetes? Researchers at the University of Bristol’s Department of Engineering Mathematics think so. Led by Ph.D. student Harry Emerson, who himself has type 1 diabetes, the team has been exploring the potential of artificial intelligence (AI) in monitoring blood sugar levels and optimizing insulin dosing.
Type 1 diabetes is a chronic condition characterized by high blood sugar levels due to the body’s inability to produce insulin. People with type 1 diabetes need to closely monitor their blood sugar levels throughout the day and inject insulin when necessary. This process typically involves obtaining blood samples from the fingers, manually recording the results, and adjusting insulin doses accordingly.
While some individuals have access to advanced tools like continuous glucose monitors (CGMs) that can provide real-time blood sugar readings, these devices still rely on hard-coded rules and do not offer flexibility in insulin dosing. Additionally, forgetting to carry the CGM monitor can hinder the transmission of data.
Emerson and his team set out to leverage machine learning to address these challenges. Collaborating with the University Hospital Southampton, they trained an AI algorithm using simulated patients generated by the UVA/Padova Type 1 Diabetes Simulator. This software is approved by the FDA and serves as an alternative to animal testing.
The AI was trained on seven months’ worth of data and learned to determine the appropriate amount of insulin to administer in different real-life scenarios. The researchers focused on reinforcement learning, which involves trial and error to optimize the algorithm’s performance. After offline training, the AI was able to autonomously decide insulin dosing for the virtual patients.
According to Emerson, the AI algorithm performed exceptionally well in the simulator, surpassing current state-of-the-art controllers for diabetes management. It even demonstrated improved performance when faced with missing or inaccurate data.
This research holds promise for revolutionizing type 1 diabetes management. By utilizing machine learning, individuals with type 1 diabetes may have access to more efficient and flexible tools for monitoring their blood sugar levels and administering insulin. Although further studies and real-life testing are necessary, the potential benefits are significant.
It’s important to note that this technology is still in its early stages and requires further development and validation. However, the use of AI in managing type 1 diabetes could potentially enhance the quality of life for millions of individuals worldwide. With advancements in technology and ongoing research, the future of diabetes management may soon be shaped by the power of machine learning.
