Our research focuses on the development of artificial intelligence methods for personalized diabetes management. In particular, we investigate data-driven approaches for insulin treatment optimization and blood glucose prediction, with the aim of supporting therapy adaptation and improving glycaemic control in people with diabetes.
A major focus of our work is the development of the Adaptive Basal-Bolus Advisor (ABBA), a reinforcement learning-based decision-support system for personalized adjustment of basal and bolus insulin in people treated with multiple daily injections. ABBA was initially developed for use with continuous glucose monitoring and insulin-pump therapy and was later extended to support sparse glucose measurements from self-monitoring of blood glucose and flash glucose monitoring, as well as insulin-pen users. ABBA has been validated in silico in type 1 and type 2 diabetes and has also been evaluated in a first-in-human feasibility study in adults with type 1 diabetes on multiple daily injections. The current results support its further clinical evaluation within the MELISSA project.
In parallel, we develop blood glucose prediction models to support proactive diabetes self-management and treatment optimization. Our recent work includes transformer-based models that combine long-term historical glucose, insulin and carbohydrate data with demographic and clinical information. These models are designed to handle incomplete real-world data, improve forecasting accuracy across heterogeneous populations, and provide early warning capabilities for hypo- and hyperglycaemic events.
Research funded by the European Commission and the Swiss Confederation – State Secretariat for Education, Research and Innovation (SERI) through the MELISSA project (Grant 101057730).
