AI语音驱动放射科口试模拟器用于住院医师培训的可行性初步研究
本研究开发了基于AI语音驱动的放射科口试模拟器RadBoardsAI,整合Whisper、GPT-4o与语音合成实现实时语音交互。8名住院医师完成基线,4名完成干预后评估,干预后信心、准备度提高,压力下降。初步证实该工具在放射科培训中具有技术可行性与可接受性。
要点
- 该AI模拟器可提供标准化口试练习,缓解传统模拟口试对师资资源的依赖
- 住院医师干预后信心和准备度提高、压力下降,提示可用于日常备考训练
- 图像标注和病例扩展是使用者反馈的首要改进方向
核心结论
- RadBoardsAI在放射科住院医师中展示出技术可行性与可接受性,可提供标准化口试准备并缓解师资限制
研究设计与样本
- 单中心IRB批准的试点研究,研究周期为2025年5月至2026年6月,采用前后调查设计
- 开发了基于网络平台的RadBoardsAI,集成实时语音交互(Whisper、GPT-4o、语音合成)与确定性考官约束
- 纳入PGY-2至PGY-5放射科住院医师,8人完成基线评估,4人(50%)完成干预后评估
- 采用系统可用性量表(SUS)评估可用性,并通过前后调查比较信心、准备度与压力
诊断标准与关键数值
- 基线信心评分为2.57±0.98,准备度评分为2.14±0.90,压力评分为3.14±0.90
- 干预后信心评分为3.50,标准化均差(SMD)为+0.95;准备度评分为3.50,SMD为+1.51
- 干预后压力评分降至1.50,SMD为-1.83
- 系统可用性量表(SUS)平均得分为70.0±15.1,评级为「Good」
临床应用与阅片要点
- 可用于住院医师日常口试模拟训练,弥补传统模拟口试对大量师资时间的依赖
- 75%的住院医师既往无模拟口试经验,该工具可作为首次接触口试流程的低门槛途径
局限性
- 干预后仅4人完成评估,样本量极小,结果仅为初步可行性数据
- 单中心试点设计,结果外推至其他培训项目需谨慎
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
BACKGROUND: The American Board of Radiology is transitioning radiology certifying examinations to a virtual oral format in 2028, creating need for scalable preparation tools, as traditional mock orals require significant faculty resources.
PURPOSE: To develop and evaluate feasibility, usability, and educational value of an AI-powered voice-driven oral board examination simulator.
METHODS: In this IRB-approved single-institution pilot (May 2025-June 2026), we developed RadBoardsAI, a web-based platform with real-time voice interaction (Whisper, GPT-4o, text-to-speech), deterministic examiner constraints, and ABR sample cases. Residents (PGY-2-PGY-5) completed simulated examinations and pre/post surveys.
RESULTS: Eight completed baselines; four (50%) post-intervention; 75% had no prior mock oral experience. Baseline confidence (2.57 ± 0.98) and preparedness (2.14 ± 0.90) were low, with moderate stress (3.14 ± 0.90). Post-intervention (n = 4) showed improved confidence (3.50; SMD = + 0.95) and preparedness (3.50; SMD = + 1.51), decreased stress (1.50; SMD = -1.83), and mean SUS 70.0 ± 15.1 (Good). Qualitative feedback identified image labeling and case expansion as priorities.
CONCLUSION: In this pilot feasibility study, RadBoardsAI demonstrates technical feasibility and acceptability among radiology residents, addressing faculty constraints while providing standardized oral board preparation.
引用
- Reddy, Brofman, Ott, et al. An AI-Powered Voice-Driven Oral Board Examination Simulator for Radiology Training: A Pilot Feasibility Study. Acad Radiol. 2026. PMID: 42810950. doi:10.1016/j.acra.2026.09.018
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。