文献速递肿瘤 PET/CT

人工智能辅助PET成像预测乳腺癌新辅助化疗反应:系统综述与荟萃分析

Artificial intelligence-assisted PET imaging for predicting neoadjuvant chemotherapy response in breast cancer: a systematic review and meta-analysis.
PubMed 文献 · Eur J Nucl Med Mol Imaging · 2025-09-17

该系统综述与荟萃分析纳入18项研究,其中11项研究共907例患者进入荟萃分析,评估人工智能辅助PET成像预测乳腺癌新辅助化疗反应的效能,合并AUC为0.80(95%CI:0.77-0.84);研究间异质性显著,患者样本量和模型整合临床数据是异质性的显著来源。

要点

  • 11项研究共907例患者进入荟萃分析,合并AUC为0.80(95%CI:0.77-0.84)。
  • 敏感度异质性I²=79.65%(95%CI:74.69-84.60),特异度异质性I²=86.62%(95%CI:83.73-89.51)。
  • 患者样本量与模型整合临床数据是异质性的显著来源。

核心结论

  • 基于PET成像的人工智能在预测乳腺癌新辅助化疗反应方面显示出有前景的准确性和临床使用潜力。
  • 其临床实施受到方法学差异、数据集小、缺乏外部验证和可解释性有限的挑战。
  • 未来研究应优先提升数据集质量并整合可解释人工智能(XAI),以促进人工智能在临床实践中的更广泛采用。

研究设计与样本

  • 系统检索Ovid MEDLINE、Ovid Embase、Cochrane、Web of Science和IEEE Xplore数据库,覆盖截至2025年6月26日的发表文献。
  • 18项研究符合系统综述纳入标准,其中11项研究(选择907例患者)纳入荟萃分析。
  • 提取二分类诊断准确性数据进行荟萃分析,以曲线下面积(AUC)为主要结局,并进行亚组分析和荟萃回归分析探索异质性来源。

主要发现

  • 荟萃分析合并AUC为0.80(95%CI:0.77-0.84)。
  • 敏感度异质性I²=79.65%(95%CI:74.69-84.60),特异度异质性I²=86.62%(95%CI:83.73-89.51)。
  • 荟萃回归分析显示患者样本量和模型整合临床数据是异质性的显著贡献因素。

关键数值与推荐

  • 合并AUC为0.80(95%CI:0.77-0.84),来自11项研究共907例患者。
  • 敏感度I²=79.65%(95%CI:74.69-84.60),特异度I²=86.62%(95%CI:83.73-89.51)。

临床怎么用

  • 原文未据此给出推荐。

局限性

  • 研究间存在显著异质性,敏感度I²=79.65%(95%CI:74.69-84.60),特异度I²=86.62%(95%CI:83.73-89.51)。
  • 临床实施受到方法学差异的挑战。
  • 数据集小。
  • 缺乏外部验证。
  • 可解释性有限。

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)。

原文

PURPOSE: This study aims to evaluate the performance of artificial intelligence (AI)-assisted PET imaging in predicting neoadjuvant chemotherapy (NAC) response in breast cancer patients. METHODS: The Ovid MEDLINE, Ovid Embase, Cochrane, Web of Science, and IEEE Xplore databases were systematically searched for studies utilizing AI algorithms in PET imaging for predicting responses to NAC in breast cancer, covering publications up to June 26, 2025. Binary diagnostic accuracy data were extracted for meta-analysis, with the area under the curve (AUC) serving as the primary outcome. Subgroup analyses and meta-regression analyses were conducted to explore potential sources of heterogeneity. RESULTS: Eighteen studies were eligible for systematic review, and eleven studies that selected 907 patients were included in the meta-analysis, with a pooled AUC of 0.80 (95% confidence interval [CI]: 0.77-0.84). However, significant heterogeneity was observed among the studies, with a I² of 79.65% (95% CI: 74.69-84.60) for sensitivity and 86.62% (95% CI: 83.73-89.51) for specificity. Meta-regression analyses revealed that patient sample size and the integration of clinical data in the models were significant contributors to heterogeneity. CONCLUSIONS: The use of AI in predicting treatment response to NAC in breast cancer based on PET imaging demonstrated promising accuracy and potential for clinical use. But its clinical implementation is challenged by methodological variability, small datasets, lack of external validation and limited interpretability. Future research should prioritize the improvement of dataset quality and the integration of explainable AI (XAI) to facilitate the broader adoption of AI in clinical practice.

来源:Eur J Nucl Med Mol Imaging,查看原文。版权归原作者所有。

引用

  1. Chen Y, Sun Y, Chen Y, et al. Artificial intelligence-assisted PET imaging for predicting neoadjuvant chemotherapy response in breast cancer: a systematic review and meta-analysis. Eur J Nucl Med Mol Imaging. 2025. PMID: 40960690. doi:10.1007/s00259-025-07536-0

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。