Improving Detection of Type-1 Diabetes Adverse Events Using GRU Networks

Recent advancements in blood glucose monitoring methods, such as Continuous Glucose Monitoring (CGM) and Continuous Subcutaneous Insulin Infusion (CSII), have significantly enhanced diabetes management, improving quality of life and reducing complications for persons affected by Type-1 Diabetes. Accurately predicting blood glucose levels and adverse events (i.e. hypoglycemia and hyperglycemia) is challenging due to cost, difficulty in collecting patient-specific data, complex interplay of glucose-regulating mechanisms, and variability in patient response to insulin. In this paper, we investigate the use and efficacy of Gated Recurrent Units (GRUs) Networks for glucose level prediction and adverse events detection in a self-supervised setting. Experimental results on benchmark datasets show an improvement in sensitivity to critical events.