Subcutaneous insulin administration by deep reinforcement learning for blood glucose level control of type-2 diabetic patients (submited for publication)

In this study, normalized advantage function (NAF) algorithm has been applied as a model-free reinforcement learning method to regulate the blood glucose level of type-two diabetic patients through subcutaneous injection. The algorithm has been designed and developed in a model-free approach in order to avoid additional inaccuracies and parameter uncertainty introduced by the mathematical models of the glucoregulatory system. Insulin dose levels which are stated directly in clinical language constitute the actions. Glucose level and effective cumulative insulin in each time frame together shape the environment’s states. Moreover, the simulation model takes into account delayed effects of the absorption dynamics of the injected insulin in addition to the glucose-insulin dynamic system. This gives a more complete and more realistic model than the previously studied models for designing controllers. Also, a simple but practical reward function is developed to be used with the NAF algorithm in order to correct the glucose level and maintain it in the desired range. Its performance was then assessed using simulations of three virtual patients to ensure reliability and robustness of the results. NAF has proved a promising control approach, able to successfully regulate and significantly reduce the fluctuation of the blood glucose without meal announcements.