医学教育管理 ›› 2026, Vol. 12 ›› Issue (5): 709-717.doi: 10.3969/j.issn.2096-045X.2026.05.020

• 专  论 • 上一篇    

人工智能在医学影像教育中的应用价值、风险与治理:一项系统评价

  

  1. 1.上海交通大学医学院附属瑞金医院脑病中心,上海 200025; 2.上海交通大学医学院医学人工智能研究院,上海 200025
  • 收稿日期:2026-04-20 修回日期:2026-05-21 出版日期:2026-11-13 发布日期:2026-11-13

Artificial intelligence in medical imaging education: a systematic review of value, risks, and governance

  1. 1. Clinical Neuroscience Center, Affiliated Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, 200025, China; 2. Institute of Medical Artificial Intelligence, Shanghai Jiao Tong University School of Medicine, Shanghai 200025, China
  • Received:2026-04-20 Revised:2026-05-21 Online:2026-11-13 Published:2026-11-13

摘要:  目的 系统评价2021—2026年人工智能(artificial intelligence,AI)在医学影像教育中的应用形态、教学价值、风险与治理要求。方法 按PRISMA原则检索PubMed、Embase、ERIC及CNKI,限定2021年1月1日—2026年3月12日发表的中英文同行评议原始研究。由2名研究者独立筛选并提取数据,根据研究设计选用偏倚风险评价工具,采用叙述性综合。结果 最初检索到6 017条记录,去重后6 014条进入题名/摘要筛选,纳入11项原始研究,每项研究样本量多在5~60人之间。AI在医学影像教育中的证据主要集中于超声训练(n=4)、影像判读、本科教学及AI素养培养。多数研究报告了短期知识/技能改善或满意度提升,但证据停留于柯氏评估1~2级,6项研究评为高偏倚风险。结论 AI可增强教学反馈和训练过程的标准化与一致性,但对复杂临床情境中的实践指导支持仍较有限。当前最可执行的路径是在窄任务、高反馈需求场景中试点AI,同步建设AI素养课程与治理框架。高质量多中心研究是下一阶段的优先需求。

关键词:

"> 人工智能|医学影像教育|放射学教育|医学生|生成式人工智能|系统评价

Abstract:

 Methods To systematically evaluate the application modalities, educational value, risks, and governance requirements of artificial intelligence (AI) in medical imaging education from 2021 to 2026. Methods  This review followed PRISMA principles and searched PubMed, Embase, ERIC, and CNKI for peer-reviewed original research published in English or Chinese between January 1, 2021, and March 12, 2026. Two reviewers independently screened and extracted data. Risk of bias was assessed using tools appropriate to the study design, and a narrative synthesis was conducted. Results  The initial search yielded 6,017 records. After deduplication, 6,014 records underwent title/abstract screening, and 11 original studies were included. The sample size in each study ranged from 5 to 60 participants. The evidence on AI in medical imaging education was mainly concentrated in ultrasound training (n=4), image interpretation, undergraduate teaching, and AI literacy development. Most studies reported short-term knowledge/skill improvement or increased satisfaction, but the evidence remained at Kirkpatrick levels 1–2, with six studies rated at high risk of bias.  Conclusion  AI can enhance the standardization and consistency of instructional feedback and training processes, but its support for practical guidance in complex clinical contexts remains limited. The most actionable path now is to pilot AI in narrow-task, high-feedback-demand scenarios while concurrently developing AI literacy curricula and governance frameworks. High-quality multicenter studies are a priority for the next stage.

Key words:

artificial intelligence|medical imaging educationv, radiology education| medical students| generative artificial intelligence| systematic review