文献速递乳腺技术与安全MR

T1 mapping 对乳腺癌多参数 MRI 影像组学预测生物学特征的初步价值

Preliminary Study of the Added Value of T1 Mapping in Multiparametric MRI Radiomics for Noninvasive Prediction of Biomarkers and Pathological Features in Breast Cancer.
PubMed 文献 · Acad Radiol · 2026-09-02

本研究纳入 101 例乳腺癌患者,比较含与不含 T1 mapping 的多参数 MRI 影像组学模型对 ER、PR、HER2、Ki-67、组织学分级及腋窝淋巴结状态的预测效能。结果显示整合临床与影像组学的综合模型预测效能最高,测试集 AUC 为 0.862–0.990。研究结论为初步结果,需更大规模多中心验证。

要点

  • 整合临床与影像组学的综合模型对 ER、PR、HER2 预测 AUC 均超过 0.94,优于单纯影像组学模型。
  • T1 mapping 加入常规多参数 MRI 有助提升乳腺癌生物学特征的无创预测能力。
  • SHAP 分析可识别对组织学分级和淋巴结状态贡献最大的影像组学特征,辅助理解模型决策。

核心结论

  • 整合 T1 mapping、常规多参数 MRI 影像组学与临床数据的综合模型可对乳腺癌生物学标志物及病理特征进行无创预测。
  • 综合模型测试集 AUC 分别为 ER 0.944、PR 0.990、HER2 0.956、Ki-67 0.862、组织学分级 0.931、腋窝淋巴结状态 0.912。
  • 该结果为初步研究结果,需在更大规模多中心队列中验证后方可向临床转化。

研究设计与样本

  • 回顾性初步研究,共纳入 101 例乳腺癌患者。
  • 所有患者均行多参数 MRI 检查,序列包括 T1 mapping、T2 加权成像、扩散加权成像及动态对比增强。
  • 图像通过 Elastix 配准至 DCE-MRI,利用 PyRadiomics 提取影像组学特征并标准化。
  • 构建五类模型:常规 mpMRI、含 T1 mapping 的非对比 mpMRI、综合 mpMRI、临床模型及整合临床影像组学模型。

关键影像表现/征象

  • SHAP 分析识别出对组织学分级和腋窝淋巴结状态预测贡献最大的关键影像组学特征。
  • 列线图可对乳腺癌个体化风险进行直观评估。

诊断标准与关键数值

  • ER 阳性/阴性分布为 85/16,PR 阳性/阴性分布为 75/26,HER2 阳性/阴性分布为 56/45。
  • Ki-67 高/低表达分布为 57/34,组织学分级高/低为 28/68,腋窝淋巴结阳性/阴性为 39/62。
  • 综合临床影像组学模型测试集 AUC:ER 0.944、PR 0.990、HER2 0.956、Ki-67 0.862、组织学分级 0.931、腋窝淋巴结状态 0.912。
  • 特征筛选采用方差分析与最小绝对收缩和选择算子回归,95% 置信区间由测试集 1000 次自举重采样估计。

临床应用与阅片要点

  • 阅片时可关注 T1 mapping 定量参数所反映的组织特征,其与常规序列结合可能提高对乳腺癌分子分型的判断信心。
  • 综合模型列线图可为个体化风险评估提供参考,但当前阶段不宜单独作为临床决策依据。
  • SHAP 所识别的关键特征对应的图像区域可作为阅片时重点观察的结构或功能区域。

局限性

  • 研究为初步探索性工作,样本量较小。
  • 需在更大规模多中心队列中验证后方可考虑临床转化。

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

RATIONALE AND OBJECTIVES: This study presents a preliminary effort to quantify the value of T1 mapping by comparing radiomics models from conventional multiparametric magnetic resonance imaging (mpMRI) with and without T1 mapping to improve noninvasive breast cancer biological assessment.

MATERIALS AND METHODS: All enrolled patients underwent mpMRI examinations, including T1 mapping, T2-weighted imaging, diffusion-weighted imaging, and dynamic contrast-enhanced (DCE) sequences. Images were registered to DCE-MRI using Elastix. Radiomics features were extracted via PyRadiomics and normalized. Five models were built and compared: conventional mpMRI, non‑contrast mpMRI with T1 mapping, comprehensive mpMRI, clinical, and integrated clinicoradiologic models. Feature selection was conducted using analysis of variance and least absolute shrinkage and selection operator regression. Model performance was evaluated using the area under the curve (AUC), with 95% confidence intervals estimated via 1000 bootstrap resamples of the test set. Model interpretability was assessed using SHapley Additive exPlanations (SHAP) analysis and nomograms.

RESULTS: A total of 101 breast cancer patients were included (estrogen receptor [ER]+/-: 85/16; progesterone receptor [PR]+/-: 75/26; human epidermal growth factor receptor 2 [HER2]+/-: 56/45; Ki-67 high/low: 57/34; histological grade high/low: 28/68; axillary lymph node [ALN]+/-: 39/62). The integrated clinicoradiologic model achieved the highest predictive performance, with test AUCs of 0.944 for ER, 0.990 for PR, 0.956 for HER2, 0.862 for Ki-67, 0.931 for histological grade, and 0.912 for ALN status, outperforming all ablated models. SHAP analysis identified key radiomics features for grade and nodal status, while nomograms offered intuitive individualized risk assessment.

CONCLUSION: Radiomics models incorporating T1 mapping with conventional mpMRI, particularly when combined with clinical data, provide a noninvasive approach for characterizing breast cancer biology and may aid personalized clinical management. Nevertheless, these findings remain preliminary and require validation in larger, multicenter cohorts before clinical translation.

来源:Acad Radiol,查看原文。版权归原作者所有。

引用

  1. Lian, He, Liang, et al. Preliminary Study of the Added Value of T1 Mapping in Multiparametric MRI Radiomics for Noninvasive Prediction of Biomarkers and Pathological Features in Breast Cancer. Acad Radiol. 2026. PMID: 42686493. doi:10.1016/j.acra.2026.08.083

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。