医疗数据场景下的联邦学习研究综述

A Review of Federated Learning in Medical Data Scenarios

随着人工智能与医疗大数据的融合,医疗数据隐私保护成为制约其发展的核心问题,联邦学习作为新型分布式隐私计算框架,旨在解决隐私安全于数据共享之间的矛盾。目前基于联邦学习的框架已运用于诸多医疗场景,取得较好成效,因此对联邦学习的概念、应用和未来发展的分析尤为重要。文章首先介绍联邦学习的概念、分析特点和列举现有框架;接着对联邦学习的三个应用领域,医疗影像、疾病风险预测和药物挖掘进行梳理,并介绍最新相关研究;最后从可解析性、安全性、性能效率三个角度探讨展望联邦学习未来研究方向。

With the integration of artificial intelligence and medical big data,the privacy protection of medical data has become the core problem that restricts its development.As a new distributed privacy computing framework,federated learning aims to solve the contradiction between privacy security and data sharing.At present,the framework based on federal learning has been applied to many medical scenarios and achieved good results.Therefore,it is particularly important to analyze the concept,application and future development of federal learning.This paper first introduces the concept of federated learning,the characteristics of analysis and lists the existing framework;Then it combs the three application fields of federal learning,medical imaging,disease risk prediction and drug mining,and introduces the latest relevant research;Finally,the future research directions of federated learning are discussed from the perspectives of resolvability,security and performance efficiency.