放射科住院医师动手型AI课程:三年可行性研究
本研究在单一学术机构连续三年(2023-2025)为27名放射科住院医师实施8小时动手型AI轮转课程,包含卷积、影像组学、机器学习、深度学习与模型评估等模块。结果显示课程结构与时长保持一致,所有住院医师完成评估,培训后答题正确率估计为80-90%。调查反馈总体偏正面,但技术复杂度与培训水平匹配度等维度分歧明显,多机构前瞻性验证仍属必要。
要点
- 8小时动手型AI轮转可在放射科住院医师培训中连续多年重复实施
- 培训后知识评估估计正确率约80-90%,但该数据为回顾性估计而非验证性结果
- 学员反馈建议为零基础者增加入门铺垫并强化临床应用环节
核心结论
- 结构化动手型AI课程整合入住院医师培训具备可行性与可持续性,培训后评估表现及混合偏正面的调查评分提示学员获得基础AI能力。
- 学员反馈明确指出的改进方向包括增加入门级铺垫和更多临床情境,体验式学习可能有助于连接AI理论与影像实践。
- 在推荐更广泛或规模化采用之前,需要多机构和前瞻性验证研究。
研究设计与样本
- 单中心教育创新研究,2023年至2025年连续三个年度队列,总样本量N=27,各年队列分别为9、8、10人。
- 课程为8小时互动式AI轮转,整合简短教学讲授与结构化实验模块,内容涵盖卷积、影像组学、机器学习、深度学习、模型评估、偏倚及病例应用。
- 评估设计以轮转后知识展示和定性反馈为主,未采用前后对照,轮转后书面测验覆盖核心课程领域。
- 学员感知采用9条目5点李克特量表调查(1=非常不同意,5=非常同意),辅以自由文本反馈,27名符合条件者中16人回复。
诊断标准与关键数值
- 所有三年队列(N=27)全部完成轮转及轮转后评估,课程结构与时长在各队列间保持一致。
- 轮转后总体评估表现估计为约80-90%正确率,但个体分数未集中存档,该百分比为近似回顾性估计而非验证性结果数据。
- 李克特量表调查中总体价值维度62%同意/非常同意对19%不同意/非常不同意,AI局限性认知维度62%对12%。
- 技术复杂度与培训水平匹配度维度50%受访者不同意其匹配自身训练水平,显示该维度分歧接近均分。
临床应用与阅片要点
- 该课程模式可作为放射科住院医师AI基础能力培训的可复制模板,8小时的轮转结构便于嵌入现有培训体系。
- 自由文本反馈一致建议为无AI或编程经验的学员提供入门级引导,并加大临床应用比重,实施时应据此调整教学顺序与案例选择。
- 课程中「AI局限性」相关领域评分相对正面,提示此类培训有助于帮助阅片医生理解AI辅助工具的适用边界。
局限性
- 培训后评估个体分数未集中存档,80-90%正确率为回顾性近似估计,缺乏经过验证的结局数据支持。
- 调查仅16/27名符合条件者回复,且属单一学术机构的自我报告反馈,外部推广性有限。
- 研究设计未采用教育干预的前后比较,难以量化课程带来的增量知识获益。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
RATIONALE AND OBJECTIVES: Artificial intelligence (AI) has rapidly transformed radiology practice, yet structured and practical AI education remains inconsistently integrated into radiology residency training. We developed and implemented a hands-on AI curriculum designed to integrate core computational principles with clinically relevant imaging applications. This study describes the curriculum design and evaluates its feasibility, reproducibility, and preliminary educational outcomes over three consecutive years of implementation.
MATERIALS AND METHODS: An 8-hour interactive AI rotation was incorporated into the diagnostic radiology residency curriculum at a single academic institution from 2023 through 2025. The curriculum combined brief didactic instruction with structured laboratory modules covering convolution, radiomics, machine learning, deep learning, model evaluation, bias, and case-based applications. Each year, a new cohort completed the curriculum. Cohort sizes were 9 residents in 2023, 8 in 2024, and 10 in 2025 (total N=27). As an educational innovation initiative, evaluation focused on post-rotation knowledge demonstration and qualitative feedback rather than pre-post comparison. Post-rotation knowledge was assessed with written quizzes covering core curriculum domains, and learner perceptions were subsequently evaluated using a 9-item, 5-point Likert-scale survey (1 = strongly disagree, 5 = strongly agree) administered to residents who had completed the curriculum, supplemented by free-text feedback.
RESULTS: Across three consecutive cohorts (N=27), all residents completed the rotation and post-rotation assessment, and curricular structure and duration remained consistent across cohorts, supporting feasibility and reproducibility. Estimated aggregate post-training assessment performance was approximately 80-90% correct responses, though individual scores were not centrally archived and this percentage should be interpreted as an approximate, retrospective estimate rather than validated outcome data. A survey administered to residents who had completed the curriculum (16 of 27 eligible residents responded) showed mixed to favorable perceptions across nine domains. Ratings were most positive for overall value (62% agree/strongly agree versus 19% disagree/strongly disagree) and awareness of AI limitations (62% versus 12%), but split closer to evenly for several other domains, including appropriateness of technical complexity, where 50% of respondents disagreed that complexity matched their training level. Free-text feedback consistently recommended a more introductory primer for learners without prior AI/coding exposure and greater emphasis on clinical application.
CONCLUSION: A structured, hands-on AI curriculum integrated into residency training is feasible and sustainable across independent cohorts, with post-training assessment performance and mixed-to-favorable learner survey ratings suggesting attainment of foundational AI competencies. Learner feedback also identified specific opportunities for improvement, including greater introductory scaffolding and clinical context. Experiential learning may help bridge theoretical AI principles and clinical imaging practice. Multi-institutional and prospectively validated studies are needed before broader or scalable adoption can be recommended.
引用
- Zhang, Ge. Hands-on Artificial Intelligence Education for Radiology Residents: A Three-year Feasibility and Curriculum Implementation Study. Acad Radiol. 2026. PMID: 42701044. doi:10.1016/j.acra.2026.08.095
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。