基于随机对照试验的放射诊断AI疗效评价:范围综述
本范围综述系统检索PubMed、Embase和Web of Science,纳入截至2026年3月发表的评估放射影像诊断AI工具的随机对照试验。分析显示AI工具主要作为面向临床医生的决策辅助手段部署,通常与更高的敏感度或病灶检出率及更短的图像处理时间相关。但在急诊等复杂场景中获益较小,低特异度仍是常见局限。
要点
- AI工具在标准化任务中更可能提升敏感度并缩短图像处理时间。
- 急诊等复杂场景中AI敏感度获益较小,特异度常偏低,报告时应警惕假阳性。
- 当前证据多为临床医生辅助决策模式,尚需多中心前瞻性RCT支持临床转化。
核心结论
- AI工具主要作为面向临床医生的决策辅助手段部署,在标准化任务中可能最有益,效果因临床场景而异。
研究设计与样本
- 范围综述,检索PubMed、Embase和Web of Science,纳入截至2026年3月发表的放射影像诊断AI随机对照试验。
- 采用QUADAS-2和RoB 2工具评价偏倚风险,按PRISMA-ScR框架进行描述性综合。
诊断标准与关键数值
- AI工具通常与更高的敏感度或病灶检出率相关,并缩短图像处理时间。
- 复杂场景如急诊中获益较小,低特异度是常见局限。
临床应用与阅片要点
- 作为辅助工具,AI可在标准化场景中提高影像敏感度并减少漏诊。
- 在急诊等复杂场景中应用时应谨慎,注意特异度偏低带来的假阳性风险。
局限性
- 原文指出需要更大规模多中心前瞻性RCT,并采用一致的具有临床意义的终点指标。
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
OBJECTIVE: Artificial intelligence (AI) demonstrates significant potential in medical imaging diagnosis, yet its real-world clinical value requires validation through high-quality randomized controlled trials (RCTs). Existing RCTs report heterogeneous results across settings and outcomes, motivating a scoping review to map current evidence and identify gaps.
MATERIALS AND METHODS: This scoping review mapped RCTs published up to March 2026 that evaluated AI tools for imaging-based diagnosis in radiology in clinical settings. We systematically searched PubMed, Embase, and Web of Science, screened studies using predefined eligibility criteria, and extracted study characteristics and outcomes. Risk of bias was assessed using QUADAS-2 and the RoB 2 tool, and findings were synthesized descriptively in line with PRISMA-ScR.
RESULTS: By analyzing the included RCTs, AI tools for imaging-based diagnosis in radiology were mainly deployed as clinician-facing decision aids and were generally associated with higher sensitivity or lesion detection rates and shorter image-processing time. However, the benefits were smaller in complex scenarios such as emergency care, and low specificity remained a common limitation.
CONCLUSION: Overall, AI tools for imaging-based diagnosis in radiology are currently used mainly as clinician-facing decision aids and may be most beneficial in standardized tasks, with effects varying across clinical settings. Larger multicenter prospective RCTs with consistent, clinically meaningful endpoints are needed to support robust clinical translation.
KEY POINTS: Question Is there any high-quality evidence from randomized controlled trials (RCTs) to evaluate the clinical benefits of artificial intelligence (AI) in radiological image diagnosis? Findings Through a review of nine RCTs, AI tools for imaging-based diagnosis tended to raise sensitivity, but gains were smaller in emergency care and specificity often remained low. Clinical relevance As an assistive tool, AI can improve imaging sensitivity and reduce missed diagnoses in standardized scenarios; however, further high-quality RCT evidence is still needed.
引用
- Yan, Liu, Fu, et al. Efficacy evaluation of artificial intelligence in radiological imaging diagnosis based on randomized controlled trials: a scoping review. Eur Radiol. 2026. PMID: 42406054. doi:10.1007/s00330-026-12715-0
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。