CT人工智能术前预测肝细胞癌微血管侵犯的系统综述与Meta分析
该系统综述与Meta分析纳入截至2025年2月的32项研究、3709例肝细胞癌病例,评价基于CT的人工智能模型术前检出微血管侵犯的诊断效能。内部验证集中AI的合并敏感度为0.83、特异度为0.81、AUC为0.89,放射科医师对应敏感度为0.82、特异度为0.65、AUC为0.80。现有证据提示CT-AI可能优于放射科医师,但受异质性及直接对比研究数量限制。
要点
- CT-AI预测肝细胞癌MVI的内部验证集AUC为0.89,敏感度0.83,特异度0.81。
- 放射科医师预测MVI的AUC为0.80,特异度仅0.65,AI在特异度上优势更明显。
- 现有证据异质性较大,缺乏前瞻性多中心验证,阅片报告中应谨慎解读AI结果。
核心结论
- 基于CT的AI模型在肝细胞癌术前预测微血管侵犯方面可能优于放射科医师。
- 现有证据质量受研究异质性及AI与放射科医师直接对比研究数量有限的影响。
研究设计与样本
- 系统综述与Meta分析,检索PubMed、Embase和Web of Science中截至2025年2月的研究。
- 共识别918项研究,最终纳入32项研究、3709例肝细胞癌病例。
- 采用双变量随机效应模型计算合并敏感度、特异度和AUC及95%CI。
- 研究间异质性采用I²统计量评估,方法学质量采用改良QUADAS-2工具评价。
诊断标准与关键数值
- AI内部验证集的合并敏感度为0.83(95%CI 0.79-0.87),特异度为0.81(95%CI 0.76-0.86),AUC为0.89(95%CI 0.86-0.92)。
- 放射科医师的敏感度为0.82(95%CI 0.63-0.93),特异度为0.65(95%CI 0.45-0.81),AUC为0.80(95%CI 0.77-0.84)。
临床应用与阅片要点
- AI在特异度上相对放射科医师数值更高,提示可能有助于减少MVI假阳性判断。
- 临床采用前应关注模型开发所用CT序列、ROI勾画方式及验证集来源是否与本院匹配。
局限性
- 研究间存在异质性,AI与放射科医师直接对比的研究数量有限。
- 原文指出需要前瞻性多中心研究验证其临床应用价值。
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
PURPOSE: This meta-analysis evaluates the diagnostic performance of computed tomography (CT)-based artificial intelligence (AI) models versus radiologists for preoperative microvascular invasion (MVI) detection in hepatocellular carcinoma (HCC).
METHODS: A systematic literature search was conducted in PubMed, Embase, and Web of Science to identify studies published up to February 2025 focusing on the diagnostic accuracy of CT-based AI models for the preoperative detection of MVI in HCC, compared with the diagnostic performance of radiologists. A bivariate random-effects model was employed to calculate the pooled sensitivity, specificity, and area under the curve (AUC), all presented with 95% confidence intervals (CIs). Heterogeneity among studies was assessed using the I2 statistic. The methodological quality of included studies was evaluated using a modified version of the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool.
RESULTS: Of 918 identified studies, 32 studies with 3,709 cases were included. For the internal validation set, the pooled sensitivity, specificity, and AUC for detecting MVI in HCC were 0.83 (95% CI 0.79-0.87), 0.81 (95% CI 0.76-0.86), and 0.89 (95% CI 0.86-0.92), respectively. Radiologists achieved a sensitivity of 0.82 (95% CI 0.63-0.93), specificity of 0.65 (95% CI 0.45-0.81), and AUC of 0.80 (95% CI 0.77-0.84).
CONCLUSIONS: CT-based AI may have the potential to outperform radiologists in predicting MVI in HCC. However, existing evidence is limited by study heterogeneity and limited number of the direct comparison between AI and radiologists. Prospective multicenter studies are needed to validate its clinical utility.
引用
- Fu, Zhang, Yu, et al. Computed tomography-based artificial intelligence for predicting preoperative microvascular invasion in hepatocellular carcinoma: a systematic review and meta-analysis. Radiol Med. 2026. PMID: 41505041. doi:10.1007/s11547-025-02170-0
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。