Continuous action deep reinforcement learning for propofol dosing during general anesthesia

作者：

Highlights：

• Continuous action deep RL outperforms PID control of anesthetic delivery.

• Incentivizing dose minimization improves RL agent's overall performance.

• Deep RL can produce interpretable anesthetic control policies.

• Trained agent's actions agree with real-world doses administered during surgeries.

摘要

•Continuous action deep RL outperforms PID control of anesthetic delivery.•Incentivizing dose minimization improves RL agent's overall performance.•Deep RL can produce interpretable anesthetic control policies.•Trained agent's actions agree with real-world doses administered during surgeries.

论文关键词：Anesthesia,Reinforcement learning

论文评审过程：Received 22 April 2021, Revised 26 September 2021, Accepted 23 November 2021, Available online 2 December 2021, Version of Record 9 December 2021.

论文官网地址：https://doi.org/10.1016/j.artmed.2021.102227