基于强化学习的脓毒症用药策略研究  

Research on medication strategy of sepsis based on reinforcement learning

作　　者：王德勇 张天逸[1] 程云章[1] WANG Deyong;ZHANG Tianyi;CHENG Yunzhang(Shanghai Engineering Research Center of Interventional Medical Devices,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学上海介入医疗器械工程技术研究中心,上海200093

出　　处：《生物医学工程研究》2022年第4期397-404,共8页Journal Of Biomedical Engineering Research

摘　　要：为解决难以为脓毒症患者提供针对性治疗方案的问题,本研究基于MIMIC重症监护数据,利用强化学习中的价值迭代和策略迭代方法,求解了两套静脉输液和血管升压药用药策略。结果表明,由模型所得策略在不同患者轨迹上求得的平均累计回报要明显高于临床医生的用药策略。本研究可为临床医生制定脓毒症治疗策略提供决策参考,在临床脓毒症用药指导上具有广阔应用前景。Aim to the difficulty of providing targeted treatment programs for sepsis patients,we used value iteration and strategy iteration methods in reinforcement learning to solve two sets of intravenous fluid volume and vasopressor medication strategy based on MIMIC data.The results showed that the average cumulative return obtained from the strategies based on the model was significantly higher than that obtained from the medication strategies of clinicians on different patient tracks.It can provide reference for clinicians to formulate treatment strategies for sepsis,this method has a broad application prospect in the medication guidance of clinical sepsis.

关 键 词：脓毒症 治疗策略 升压药 静脉输液 强化学习 价值迭代 策略迭代 

分 类 号：R318[医药卫生—生物医学工程] R459.7[医药卫生—基础医学] R452[自动化与计算机技术—控制理论与控制工程] TP181[自动化与计算机技术—控制科学与工程]