基于深度强化学习的数字化医院病历数据加密研究

Digital hospital medical record data encryption based on deep reinforcement learning

摘要

为了提高电子病历(electronic medical record,EMR)共享调阅安全,应对隐私泄露、数据滥用风险,研究创新地基于深度强化学习(deep reinforcement learning,DRL)与深度学习完成了病历敏感信息的命名实体识别(named entity recognition,NER),然后采用高级加密标准(advanced encryption standard,AES)设计了病历数据加密模型。实验结果表明,基于深度强化学习的命名实体识别模型最高分类准确率可达0.941、匹配准确率可达0.901、F₁值可达0.876。在电子病历库中,该方法的标准化语义等级达0.90,有利于加密算法的性能提升,基于深度强化学习的加密算法实现了最短密钥生成时间,最多耗时559.93 ms;抵抗攻击的性能显著增加。研究设计的病历数据加密模型有助于保障患者的隐私权益和医疗安全,促进医疗资源的优化配置。

Abstract

To enhance the security of electronic medical record (EMR) sharing and retrieval, and address the risks of privacy leakage and data abuse, this study innovatively utilizes deep reinforcement learning (DRL) and deep learning to complete named entity recognition (NER) of sensitive medical record information. Subsequently, an encryption model for medical record data is designed using the advanced encryption standard (AES). Experimental results show that the DRL-based NER model achieves a maximum classification accuracy of 0.941, a matching accuracy of 0.901, and an F₁ score of 0.876. In the EMR database, the standardized semantic level of this method reaches 0.90, which is conducive to improving the performance of encryption algorithms. The DRL-based encryption algorithm achieves the shortest key generation time, with a maximum time consumption of 559.93 ms; its resistance to attacks significantly increases. The designed encryption model for medical record data helps to safeguard patients′ privacy rights and medical safety, and promotes the optimal allocation of medical resources.

关键词

深度强化学习 / DDPG模型 / 电子病历 / 加密算法 / 深度学习 / 命名实体识别

Key words

deep reinforcement learning / DDPG model / electronic medical records / encryption algorithm / deep learning / identity of named entity