Detection and prediction of physical activity in type 1 diabetic subjects

The challenge of managing glycemia during and after physical activities (PA) poses difficulties for individuals with type 1 diabetes, as the impact of the activity on blood sugar levels can vary greatly depending on various factors. Therefore, this proposed research aims to create a real-time PA detection module on the T1DEXI dataset, with the objective of enhancing glucose control.The proposed research includes the model-based detection approach mainly leverages on sensor-measured glucose levels, infused insulin and carbohydrates ingested in meals, with the optional inclusion of other signals available from smartwatches. It is based on the monitoring of various statistical properties of the residuals, i.e., the difference between the glucose levels measured by the sensor and those predicted by a model. The goal is to provide a practical aid to rapidly detect PA in real-time, requiring in input the minimum amount of information possible. Furthermore, the proposed research includes the development of model-free PA detection, based on supervised and/or unsupervised data-driven methodologies, leveraging on all the available data present in the dataset. The goal is to provide a detection module, possibly able to detect and classify the PA in real-time. Finally, the proposed research aims at the development of model-based techniques for hypo/hyper glycemia detection, following PA.