基于影像的人工智能诊断胆道闭锁:系统综述与Meta分析
本系统综述与Meta分析纳入9项研究、11006例患者及11500张超声图像,评估基于影像的AI模型识别胆道闭锁的诊断效能。患者水平汇总敏感度为93.8%,特异度为93.2%,AUC为0.94;图像水平汇总敏感度为86.9%,特异度为94.3%,AUC为0.965。结论认为AI表现良好,可作为临床决策辅助工具,但需更多高质量研究验证其普适性。
要点
- AI在超声图像中识别胆道闭锁的患者水平汇总AUC达0.94,可作为辅助筛查工具。
- 患者水平敏感度93.8%高于图像水平86.9%,提示以患者为单位分析更接近临床实际。
- 图像水平特异度94.3%略高于患者水平93.2%,AI排除非胆道闭锁的效能较稳定。
核心结论
- AI在基于影像的胆道闭锁诊断中表现令人满意,应作为支持临床决策的辅助工具。
- 仍需进一步高质量研究确认其在不同人群和场景下的普适性。
研究设计与样本
- 本研究为系统综述与Meta分析,遵循PRISMA DTA指南并在PROSPERO注册。
- 检索PubMed、Web of Science、Embase和Scopus数据库,使用预设纳入标准筛选研究报告。
- 共纳入9项研究,分析11006例患者,其中2357例胆道闭锁、8649例非胆道闭锁,涉及11500张超声图像。
- 使用QUADAS-AI和QUADAS-2进行质量评估,并应用GRADE方法评价证据确定性。
诊断标准与关键数值
- 患者水平分析汇总敏感度为93.8%(95% CI:86-97.4%),特异度为93.2%(95% CI:91.8-94.4%),AUC为0.94。
- 图像水平分析汇总敏感度为86.9%(95% CI:73.7-94.1%),特异度为94.3%(95% CI:90.9-96.4%),AUC为0.965。
- 纳入研究总体质量为中至高,但数项研究在指数测试域存在高偏倚风险。
临床应用与阅片要点
- AI可作为超声诊断胆道闭锁的辅助工具,帮助提高检出率并支持临床决策。
- 阅片时可将AI提示与常规灰阶超声及肝门部纤维块等征象结合,警惕假阳性与假阴性。
- 注意患者水平与图像水平AI效能存在差异,报告中宜注明分析单位。
局限性
- 数项纳入研究在指数测试域存在高偏倚风险。
- 原文提出仍需进一步高质量研究确认AI诊断胆道闭锁的普适性。
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: Biliary atresia (BA) is a rare condition that can lead to serious health complications. Artificial Intelligence (AI)- based medical imaging has shown potential to improve the detection of BA diagnosis, offering increased accuracy over traditional imaging methods. This meta-analysis aimed to evaluate the diagnostic accuracy of imaging-based AI for detecting BA.
METHODS: The study followed PRISMA DTA guidelines and was registered in the PROSPERO database. The search was performed in PubMed, Web of Science, Embase, and Scopus databases for studies reporting the diagnostic accuracy of imaging-based AI models in detecting BA. Meanwhile, pre-defined eligibility criteria were used for the selection of studies. Pooled sensitivity, specificity, and area under the curve (AUC) were calculated utilizing R 4.4.2. Quality assessment was performed using the Quality Assessment of Diagnostic Accuracy Studies-AI (QUADAS-AI) criteria and the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool, and the GRADE approach was applied to evaluate the certainty of the synthesised evidence.
RESULTS: A total of nine studies were included in this meta-analysis. The overall quality of the included studies was moderate to high, although a high risk of bias was noted in the index test domain of several studies. A total of 11,006 patients were analyzed, comprising 2,357 BA cases and 8,649 non-BA cases, with 11,500 ultrasound images. Patient-level analysis revealed pooled sensitivities, specificities, and AUCs of 93.8 % (95 % CI: 86-97.4 %), 93.2 % (95 % CI: 91.8-94.4 %), and 0.94, respectively. Additionally, the pooled results of image-based analysis revealed sensitivity, specificity, and AUC of 86.9 % (95 % CI: 73.7-94.1 %), 94.3 % (95% CI: 90.9-96.4 %), and 0.965, respectively.
CONCLUSION: AI shows satisfactory performance in the imaging-based diagnosis of biliary atresia. It should be regarded as an assistive tool that supports clinical decision-making, and further high-quality studies are needed to confirm its generalizability.
引用
- Halimi, Msherghi, Nounou, et al. Artificial intelligence performance in image-based biliary atresia identification: a systematic review and meta-analysis. Eur J Radiol. 2026. PMID: 41499913. doi:10.1016/j.ejrad.2025.112651
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。