Basal Glucose Control in Type 1 Diabetes Using Deep Reinforcement Learning: An <i>In Silico</i> Validation

People with Type 1 diabetes (T1D) require regular exogenous infusion of insulin to maintain their blood glucose concentration in a therapeutically adequate target range. Although the artificial pancreas and continuous monitoring have been proven be effective achieving closed-loop control, significant challenges still remain due high complexity dynamics limitations technology. In this work, we propose novel deep reinforcement learning model for single-hormone (insulin) dual-hormone (insulin glucagon) delivery. particular, delivery strategies are developed by double Q-learning dilated recurrent neural networks. For designing testing purposes, FDA-accepted UVA/Padova simulator was employed. First, performed long-term generalized training obtain population model. Then, personalized small data-set subject-specific data. silico results show that single achieve good control when compared standard basal-bolus therapy low-glucose suspension. Specifically, adult cohort (n=10), percentage time range [70, 180] mg/dL improved from 77.6% 80.9% $85.6\%$ control. adolescent 55.5% 65.9% 78.8% all scenarios, decrease hypoglycemia observed. These use is viable approach T1D.