Joint Behavioral Cloning And Reinforcement Learning Method For Propofol And Remifentanil Infusion In Anesthesia

Total intravenous anesthesia using propofol and remifentanil is widely used. However, a rapid elimination property of the control of the drug infusion rate leads to problems. Recently, it has been shown that machine learning algorithms are able to more accurately predict drug effects than traditional prediction models. Based on this trend, this paper proposes a machine learning method that can simulate the transition dynamics of patient data during anesthesia to configure the interactive environment. In addition, we propose deep reinforcement learning with behavior cloning-based initialization to train the propofol and remifentanil control policies. As shown in the data-intensive performance evaluation, the proposed method achieves a desirable performance in terms of average bispectral index and blood pressure during the surgery simulation.