医学人工智能公共服务平台研究与建设

Research and construction of medical artificial intelligence public service platform

  • 摘要:
    背景 以深度学习、卷积神经网络、大模型等技术为代表的新一代人工智能发展迅猛,深入影响医疗卫生领域。
    目的 为解决医疗健康行业人工智能发展所面临的数据资源短缺、算力条件不足、协同支撑匮乏等问题,研究建设全新的医学人工智能公共服务平台。
    方法 依托国家工信部2020年“产业技术基础公共服务平台”项目,针对医疗AI的能力生成与服务支撑等需求,基于“多域融合+云边协同”理念,提出“云-网-边-端”先进架构,着力研究训推异构兼容和智能联邦学习等技术。
    结果 建成了一个主要采用国产软硬件设备,面向医疗健康行业,集人工智能数据治理、模型训练、测试验证及推理应用等核心功能于一体的大型公共服务平台,实现了医疗数据、计算资源和应用服务在云、边、端之间的优化配置与高效协同。在全国240家基层军地医疗机构应用推广医学人工智能筛查与辅助诊断产品,预计服务上千万人次。
    结论 医学人工智能公共服务平台的研究与建设,有效推动了医疗数据汇聚、基础算子研制、高端算力构建、开源模型研究以及医学人工智能系统应用,增强自主可控、自主创新的实力,未来将有力推进我国智慧医疗建设的可持续发展。

     

    Abstract:
    Background The new generation of artificial intelligence, represented by deep learning, convolutional neural networks, large models and other technologies, has developed rapidly and deeply affected the medical and health field.
    Objective To establish a public service platform for medical artificial intelligence, so as to solve the problems of shortage of data resources, insufficient computational power and lack of collaborative support for the development of artificial intelligence in the healthcare industry.
    Methods Based on the project of "Industrial Technology Foundation Public Service Platform" launched by the Ministry of Industry and Information Technology in 2020, with the need for medical AI capability generation and service support, and the concept of "multi-domain convergence + cloud-edge collaboration", the advanced theory of "cloud-network-edge-end" was put forward, and technologies such as training and inference heterogeneous compatibility and intelligent federal learning were investigated.
    Results A large-scale public service platform, which mainly adopted domestic software and hardware equipment and was oriented to the medical and health industry, had been established successfully to provide core functions such as artificial intelligence data governance, model training, test validation and reasoning application, the optimized configuration and efficient collaboration of medical data, computing resources and application services among cloud, edge and end were realized. The application and promotion of medical AI screening and auxiliary diagnosis products in 240 grass-roots military and civilian medical institutions across the country is expected to serve tens of millions of people.
    Conclusion The public service platform effectively promotes the convergence of medical data, accelerates the construction of independent, self-supporting, heterogeneous and compatible computing capacity services, and contributes to the research and development of medical AI systems and the sustainable development of China's smart medical construction.

     

/

返回文章
返回