针对医学人工智能产品的对抗测试技术平台设计

Design of adversarial testing technology platform for medical artificial intelligence products

目的:设计针对医学人工智能产品的对抗测试技术平台,以为医学人工智能产品提供对抗测试方法、通用性能评价和不良事件监控。方法:该平台基于浏览器/服务器(Browser/Server,B/S)架构设计,采用MySQL数据库进行搭建。该平台主要包括数据集管理、人工智能产品模型管理和人工智能性能监控3个模块,其中数据集管理模块和人工智能产品模型管理模块均采用React框架进行搭建,人工智能性能监控模块基于Nodejs框架设计,包括对抗测试技术和性能动态追踪与风险预警2个子模块。结果:该平台可实现对不同模态的人工智能产品的对抗测试、通用性能评价和不良事件监控,能够为医学人工智能产品提供全方位的评价与分析。结论:该平台解决了目前医学人工智能产品检测薄弱的问题,能够满足医学人工智能产品检测需求和监管需求。

Objective To design an adversarial testing technology platform to realize adversarial testing, general performance evaluation and adverse event monitoring for medical artificial intelligence(AI) products. Methods An adversarial testing technology platform was developed based on Browser/Server(B/S) architecture and MySQL database. There were three modules for data set management, AI product model management and AI performance monitoring involved in the platform.The dataset management module and AI product model management module were built using React framework, and the AI performance monitoring module was designed based on Nodejs framework, which included 2 sub-modules for dynamic tracking of adversarial testing technology and performance and risk warning. Results The platform developed enabled adversarial testing, general performance evaluation and adverse event monitoring of AI products of different modalities, and realized comprehensive evaluation and analysis of medical AI products. Conclusion The platform developed solves the problem in medical AI product testing and meets their demands for testing and supervision.