文献速递腹部技术与安全USMRCT

人工智能在胆囊与胆管结石影像诊断中的系统综述

Artificial intelligence in the imaging diagnosis of gallbladder and bile duct stones: a systematic review.
PubMed 文献 · Abdom Radiol (NY) · 2026-06-29

本系统综述纳入13项研究(超过7700例患者、超过130000张图像),评估人工智能在胆囊及胆管结石影像诊断中的表现。由于临床与方法学异质性大,未行荟萃分析而作定性综合。人工智能模型以超声卷积神经网络为主,准确率71.5%–99.63%,AUC 0.79–0.99,但证据确定性极低。

要点

  • AI对胆囊与胆管结石影像诊断准确率71.5%–99.63%,AUC 0.79–0.99,表现有潜力。
  • MRCP上胆总管结石检测与多分类胆囊疾病分类性能最高。

核心结论

  • AI在胆囊与胆管结石影像诊断中表现出有前景的诊断性能。
  • 方法学局限和缺乏稳健外部验证使其尚不能常规临床采用。
  • 需要前瞻性、多中心、真实世界验证研究。

研究设计与样本

  • 系统综述遵循PRISMA 2020指南,检索7个主要数据库,无语言或日期限制。
  • 纳入13项研究,涉及超过7700例患者、超过130000张图像。
  • 使用QUADAS-2评估偏倚风险,GRADE评估证据确定性,采用定性综合。

关键影像表现/征象

  • AI模型以卷积神经网络为主,主要应用于超声图像。
  • 多分类胆囊疾病分类与MRCP上胆总管结石检测的性能最高。

诊断标准与关键数值

  • AI模型报告准确率范围为71.5%–99.63%。
  • AI模型报告AUC范围为0.79–0.99。

临床应用与阅片要点

  • AI可用于辅助超声与MRCP图像中胆囊及胆管结石的检测与分类。
  • 阅片时应关注多分类胆囊疾病与MRCP胆总管结石场景,AI表现相对较高。
  • 当前证据不足以支持常规临床采用,应用时应结合常规影像诊断并保持谨慎。

局限性

  • 纳入研究临床与方法学异质性大,未能进行荟萃分析。
  • 缺乏稳健的外部验证,研究整体方法学局限明显。

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: Gallbladder and bile duct stones (cholelithiasis and choledocholithiasis) represent a major global health burden. Conventional imaging modalities have well-recognised limitations in sensitivity, operator dependence, and accessibility. Artificial intelligence (AI) has emerged as a potential adjunct to improve diagnostic accuracy, yet the evidence base remains unsynthesised.

METHODS: A systematic review was conducted in accordance with PRISMA 2020 guidelines. Seven major databases were searched without language or date restrictions. Studies evaluating any AI algorithm for the detection, segmentation, or classification of gallbladder or bile duct stones on medical images were included. Risk of bias was assessed with QUADAS-2 and evidence certainty with GRADE and a Qualitative synthesis was performed.

RESULTS: Thirteen studies (n > 7,700 patients, > 130,000 images) met inclusion criteria. Owing to substantial clinical and methodological heterogeneity, meta-analysis was not performed and findings were synthesized qualitatively. AI models, predominantly convolutional neural networks applied to ultrasound reported accuracies of 71.5-99.63% and AUC values of 0.79-0.99. Performance was highest for multi-class gallbladder disease classification and choledocholithiasis detection on MRCP. Twelve of 13 studies (92%) carried high risk of bias; overall certainty of evidence was rated very low.

CONCLUSION: AI demonstrates promising diagnostic performance for biliary stones; however, methodological limitations and lack of robust external validation preclude routine clinical adoption. Prospective, multicenter studies with real-world validation are urgently required.

来源:Abdom Radiol (NY),查看原文。版权归原作者所有。

引用

  1. Hajihashemi, Kalantary, Elhaie, et al. Artificial intelligence in the imaging diagnosis of gallbladder and bile duct stones: a systematic review. Abdom Radiol (NY). 2026. PMID: 42371021. doi:10.1007/s00261-026-05648-1

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。