基于DBT瘤内异质性生境影像组学预测HER2状态
本探索性研究纳入444例浸润性乳腺癌患者,基于数字乳腺断层摄影(DBT)提取生境影像组学特征量化瘤内异质性,用于术前无创预测HER2状态。采用多层感知机和逻辑回归构建Intra、ITH及联合模型。结果显示联合模型区分HER2阳性、HER2低表达和HER2零表达效果最佳,ITH模型在区分HER2阳性与阴性时AUC达0.934。
要点
- 生境影像组学可量化DBT上的瘤内异质性,辅助术前HER2状态评估。
- 联合模型对HER2阳性/低表达/零表达三类区分能力最优。
- 区分HER2阳性与阴性时,ITH模型测试集AUC为0.934。
核心结论
- 联合模型在区分HER2阳性、HER2低表达和HER2零表达病例时表现最佳。
- 基于生境的影像组学可量化瘤内异质性,有助于预测HER2状态。
研究设计与样本
- 研究类型为探索性研究,纳入444例浸润性乳腺癌患者,数据采集时间为2018年1月至2025年6月。
- 患者分为HER2阳性、HER2低表达、HER2零表达三组。
- 影像来源为数字乳腺断层摄影(DBT),经影像组学特征提取与降维后,采用多层感知机(MLP)和逻辑回归(LR)分类器构建Intra、ITH及联合模型。
- 共进行两项分类任务:任务1为区分HER2阳性与HER2阴性,任务2为区分HER2低表达与HER2零表达。
诊断标准与关键数值
- 任务1测试集中,ITH模型AUC为0.934(95%CI:0.881~0.988)。
- 任务1测试集中,联合模型AUC为0.940(95%CI:0.894~0.986),数值上高于ITH模型。
- 任务2测试集中,联合模型结果最优,AUC为0.849(95%CI:0.740~0.958)。
- 任务2中ITH模型表现与传统Intra模型相当。
临床应用与阅片要点
- DBT生境影像组学有望在术前无创辅助HER2状态分型,指导个体化治疗决策。
- 联合Intra与ITH特征的模型在临床阅片场景中具有更好的综合判别价值。
局限性
- 本研究为探索性设计,结论需进一步独立验证。
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
UNLABELLED: RATIONALE AND OBJECTIVE: This exploratory study aims to develop and validate a model on Digital Breast Tomosynthesis (DBT), using habitat‑based radiomics to quantify Intratumoral Heterogeneity (ITH) for non-invasive pre-operative prediction of HER2 status in invasive breast cancer.
MATERIALS AND METHODS: 444 people with invasive breast cancer were studied (2018.1-2025.6). They were classified as HER2-positive, HER2-low, or HER2-zero. After radiomics feature extraction and dimensionality reduction, we developed Intra, ITH, and combined models using multilayer perceptron (MLP) and logistic regression (LR) classifiers. Two classification tasks were performed: Task 1 is to differentiate HER2-positive versus HER2-negative individuals. Task 2, which is to differentiate individuals with HER2-low versus HER2-zero status.
RESULTS: The ITH model achieved good performance on the test set in Task 1(AUC, 0.934; 95% CI: 0.881, 0.988), while the combined model achieved a numerically higher AUC of 0.940 (95% CI: 0.894, 0.986). In Task 2, the ITH model performed comparably to the conventional Intra model, whereas the combined model produced the most favorable results on the test set (AUC, 0.849; 95% CI: 0.740, 0.958).
CONCLUSIONS: The combined model performed best at discriminating HER2-positive, HER2-low, and HER2-zero cases. Habitat-based radiomics quantifying ITH, thereby facilitating the prediction of HER2 status.
引用
- Miao, Liu, Gao, et al. Habitat-Based Radiomic Quantification of Intratumoral Heterogeneity on Digital Breast Tomosynthesis for Predicting HER2 Status in Invasive Breast Cancer. Acad Radiol. 2026. PMID: 42493348. doi:10.1016/j.acra.2026.06.063
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。