人工智能在CT中预测直肠癌患者淋巴结转移:一项Meta分析
该Meta分析纳入7项研究,评估基于CT的人工智能检测直肠癌淋巴结转移的诊断准确性;AI汇总敏感度0.84、特异度0.77、AUC 0.88,放射科医生分别为0.80、0.54、0.72;AI特异度优于放射科医生,但研究间异质性显著且外部测试有限,需进一步多中心测试后方可临床实施。
要点
- 基于CT的AI检测直肠癌淋巴结转移汇总敏感度0.84、特异度0.77,放射科医生分别为0.80、0.54。
- AI汇总AUC为0.88,放射科医生为0.72,AI特异度优于放射科医生。
- 研究间异质性显著且现有模型外部测试有限,需进一步多中心测试后方可临床实施。
核心结论
- 基于CT的AI在直肠癌淋巴结转移检测中表现出优于放射科医生的特异度。
- 研究间显著异质性和现有模型外部测试有限,需进一步多中心测试研究后方可临床实施。
研究设计与样本
- 研究类型为Meta分析,检索PubMed、Embase和Web of Science至2024年9月,纳入7项评估基于CT的AI检测直肠癌淋巴结转移诊断准确性的研究。
- 内部测试集包含23个列联表,放射科医生数据包含19个列联表。
- 采用双变量随机效应模型计算汇总敏感度和特异度并报告95%CI,采用I²统计量评估研究间异质性。
诊断标准与关键数值
- AI汇总敏感度0.84(95%CI 0.79-0.87)、特异度0.77(95%CI 0.66-0.85)、AUC 0.88(95%CI 0.16-1.00)。
- 放射科医生汇总敏感度0.80(95%CI 0.72-0.86)、特异度0.54(95%CI 0.46-0.63)、AUC 0.72(95%CI 0.11-0.98)。
- 漏斗图分析显示AI的P值为0.13、放射科医生的P值为0.40,未观察到发表偏倚。
临床应用与阅片要点
- 原文结论指出基于CT的AI在直肠癌淋巴结转移检测中特异度优于放射科医生,但需进一步多中心测试研究后方可临床实施。
局限性
- 研究间存在显著异质性。
- 现有模型的外部测试有限。
- 需进一步多中心测试研究后方可临床实施。
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
AIM: This meta-analysis evaluates the diagnostic accuracy of computed tomography (CT)-based artificial intelligence (AI) in detecting lymph node metastasis in rectal cancer patients.
MATERIALS AND METHODS: A thorough literature search was conducted in PubMed, Embase, and Web of Science to identify studies on the diagnostic accuracy of CT-based AI in detecting lymph node metastasis in rectal cancer patients, with a deadline of September 2024. The analysis assessed the performance of AI algorithms in detection through CT imaging. Pooled sensitivity and specificity were calculated using a bivariate random-effects model, reported with 95% confidence intervals. Study heterogeneity was evaluated using the I2 statistic to account for heterogeneity among the included studies.
RESULTS: Seven studies were included in this meta-analysis. The internal test set contained 23 contingency tables, while the radiologists' data included 19 contingency tables. AI's pooled sensitivity, specificity, and area under the curve (AUC) were 0.84 (0.79-0.87), 0.77 (0.66-0.85), and 0.88 (0.16-1.00), respectively. Radiologists had pooled sensitivity, specificity, and AUC of 0.80 (0.72-0.86), 0.54 (0.46-0.63), and 0.72 (0.11-0.98). Funnel plot analysis showed P values of 0.13 for AI and 0.40 for radiologists, indicating that no publication bias was observed.
CONCLUSION: CT-based AI demonstrates superior specificity over radiologists for LNM detection in rectal cancer. However, significant heterogeneity across studies and limited external tests of existing models warrant further multicenter test studies before clinical implementation.
引用
- Hou D, Her H, Han W, et al. Artificial intelligence in CT for predicting lymph node metastasis in rectal cancer patients: a meta-analysis. Clin Radiol. 2025. PMID: 41338157. doi:10.1016/j.crad.2025.107001
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。