知识引导双路径框架自动分类肝脏MRI序列
本研究开发知识引导双路径(KDP)框架,整合CNN影像特征与DICOM元数据,实现肝脏MRI 18类序列自动分类。内部测试集7,141个序列宏平均F1为97.63%,外部多中心测试集2,208个序列宏平均F1为96.76%,显著优于单纯影像模型。
要点
- KDP框架可自动识别肝脏MRI序列类型,减少手动整理错误。
- 外部多中心验证宏平均F1达96.76%,跨机构泛化能力较可靠。
- 整合DICOM头信息(采集时间、序列描述、b值)可提升分类稳健性。
核心结论
- 知识引导双路径框架对肝脏MRI序列分类达到内部宏平均F1 97.63%与外部宏平均F1 96.76%。
- 该框架显著优于单纯影像模型,外部多中心数据验证支持临床整合潜力。
研究设计与样本
- 研究类型为方法开发与多中心外部验证,内部测试集7,141个序列。
- 外部多中心测试集来自22家医院123例共2,208个序列。
- KDP框架整合CNN影像特征与DICOM元数据,通过基于规则的融合模块处理各序列类型。
- 分类覆盖18类序列,包括平扫、动态增强各期、定量图、冠状位序列等。
诊断标准与关键数值
- 内部测试集宏平均F1为97.63%(95% CI: 97.09%-98.09%),样本量7,141个序列。
- 外部多中心测试集宏平均F1为96.76%(95% CI: 95.78%-97.64%),样本量2,208个序列。
- 12个共有类别中KDP框架宏F1为96.50%(95% CI: 95.05%-97.68%),影像模型为91.54%(95% CI: 89.67%-93.12%)。
- 两模型比较McNemar检验χ²=53.6,p<0.001。
临床应用与阅片要点
- 可用于自动整理肝脏MRI多序列影像,减少PACS中序列归类的人工负担。
- 对动态增强各期与定量图的自动识别有助于后续结构化报告与流程提速。
- 实际部署前需结合本机构序列命名规范与设备参数分布评估适用性。
局限性
- 原文未报告该框架在罕见序列类型或图像质量较差情况下的分类表现。
- 自动分类错误可能导致序列标签错误传播至后续定量分析,需人工核对机制。
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: Accurate identification of liver MRI series is crucial for streamlining clinical workflows, yet current automated methods remain limited in coverage and robustness for real-world practice. This study aims to develop a clinically feasible automated classification system for liver MRI series.
METHODS: We developed a knowledge-guided dual-path (KDP) framework that integrates two complementary information sources: CNN-extracted imaging features and metadata (e.g., acquisition time, series description, b-value) from DICOM headers. A rule-based fusion module, built on predefined clinical rules, then determines the processing pathway for each sequence type based on these inputs, enabling reliable classification across all 18 series categories (including non-contrast, dynamic contrast-enhanced (DCE) phases, quantitative maps, coronal series, and others). This approach was externally validated on a multicenter test set of 2,208 series from 123 cases across 22 hospitals.
RESULTS: The proposed method demonstrated excellent performance, with macro-average F1-scores of 97.63% (95% CI: 97.09%-98.09%) on the internal test set (n = 7,141 series) and 96.76% (95% CI: 95.78%-97.64%) on the external multicentric test set (n = 2,208 series). The KDP framework significantly outperformed the image-only model, with a macro F1-score of 96.50% (95% CI: 95.05%-97.68%) versus 91.54% (95% CI: 89.67%-93.12%) on the 12 shared categories (McNemar's test, χ² = 53.6, p < 0.001).
CONCLUSION: The proposed method provides accuracy and fully automated discrimination of liver MRI series. Its validated robustness across multicentric data highlights its potential for clinical integration.
引用
- Han, Xu, Lv, et al. A knowledge-guided dual-path framework for automated liver MRI series classification. Abdom Radiol (NY). 2026. PMID: 42760361. doi:10.1007/s00261-026-05790-w
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。