Liu Jianghua, Chen Chao, Qu Youyang, Yang Shuiqiao, Xu Lei
Nanjing University of Science and Technology, China.
RMIT University, Australia.
ISA Trans. 2023 Oct;141:20-29. doi: 10.1016/j.isatra.2023.03.049. Epub 2023 Apr 6.
Powered by the rapid progress of analytics techniques and the increasing availability of healthcare data, artificial intelligence (AI) is bringing a paradigm shift to healthcare applications. AI techniques offer considerable advantages for the evaluation and assimilation of large amounts of complex healthcare data. However, to effectively use AI tools in healthcare, key issues need to be considered and several limitations must be addressed, such as privacy-preserving and authentication of the healthcare data for analysis in training and inference procedures. Although various techniques ranging from cryptographic tools to obfuscation mechanisms have been proposed to provide privacy guarantees for data in AI-based services, none of them is applicable to online AI-driven healthcare applications. For they require a heavy computational cost on protecting privacy without offering authentication services for third parties. In this paper, we present RASS, an efficient privacy-preserving and authentication scheme for securing analyzed data in an AI-driven healthcare system. The security proofs of our construction indicate that its unforgeability and multi-show unlinkability can defend against the tempering and collusion attacks respectively. Finally, we conduct sufficient efficiency analysis, and the results show that RASS achieves the above security demands without introducing complex computation and communication costs.
受分析技术的快速发展以及医疗保健数据可用性不断提高的推动，人工智能（AI）正在给医疗保健应用带来范式转变。人工智能技术在评估和处理大量复杂的医疗保健数据方面具有显著优势。然而，要在医疗保健领域有效使用人工智能工具，需要考虑关键问题并解决一些限制，例如在训练和推理过程中用于分析的医疗保健数据的隐私保护和认证。尽管已经提出了从加密工具到混淆机制等各种技术，以在基于人工智能的服务中为数据提供隐私保证，但它们都不适用于在线人工智能驱动的医疗保健应用。因为它们在保护隐私方面需要高昂的计算成本，却没有为第三方提供认证服务。在本文中，我们提出了RASS，这是一种用于在人工智能驱动的医疗保健系统中保护分析数据安全的高效隐私保护和认证方案。我们构造的安全性证明表明，其不可伪造性和多次展示不可链接性分别可以抵御篡改攻击和勾结攻击。最后，我们进行了充分的效率分析，结果表明RASS在不引入复杂计算和通信成本的情况下实现了上述安全要求。
