基于超声黏弹性成像技术构建预测乳腺癌HER2三分类表达的可解释性机器学习模型

覃坚钊 ,  黄群 ,  袁晗 ,  覃秋群 ,  廖新红 ,  高泳

中国医科大学学报 ›› 2026, Vol. 55 ›› Issue (6) : 502 -506.

PDF (1241KB)
中国医科大学学报 ›› 2026, Vol. 55 ›› Issue (6) : 502 -506. DOI: 10.12007/j.issn.0258‐4646.2026.06.004
论著

基于超声黏弹性成像技术构建预测乳腺癌HER2三分类表达的可解释性机器学习模型

作者信息 +

Interpretable machine learning model for predicting HER2 three-classification expression in breast cancer based on ultrasonic viscoelastic imaging technology

Author information +
文章历史 +
PDF (1270K)

摘要

目的 基于超声黏弹性成像技术,融合多分类机器学习和沙普利附加解释 (SHAP),构建无创预测乳腺癌患者的人表皮生长因子受体2 (HER2) 三分类模型。方法 选取2023年8月至2024年8月就诊于广西医科大学第一附属医院的150例乳腺癌患者。其中,HER2零表达44例、HER2低表达65例、HER2阳性41例。患者术前均行乳腺超声黏弹性成像检查,测量瘤内及瘤周2 mm区域的超声黏弹性参数。建模前完成数据预处理,基于5种机器学习算法构建乳腺癌HER2表达状态的三分类预测模型。用SHAP进行最优机器学习模型的可解释性分析。结果 基于超声黏弹性成像技术的多分类机器学习模型在预测乳腺癌HER2表达状态中性能良好。在验证集中,轻量梯度提升机 (LightGBM) 模型的宏平均AUC为0.827、准确率为0.875、精确率为0.860、召回率为0.863、F1-score为0.860。SHAP分析结果表明,在LightGBM模型中,乳腺癌病灶的瘤内弹性最大值 (Emax) 在预测HER2零表达和HER2阳性中贡献值最大,乳腺癌病灶的瘤周2 mm频散最大值 (Dmax_2 mm) 在预测HER2低表达中贡献值最大。结论 基于超声黏弹性成像技术构建三分类机器学习模型可在术前无创性预测乳腺癌患者HER2表达状态,结合SHAP可实现机器学习模型的可解释性分析。

Abstract

Objective To establish a three-classification preoperative predictive model to non-invasively predict the expression of human epidermal growth factor receptor 2 (HER2) in patients with breast cancer based on ultrasonic viscoelastic imaging technology combined with multi-classification machine learning and Shapley additive explanations (SHAP). Methods A total of 150 patients with breast cancer from the First Affiliated Hospital of Guangxi Medical University were enrolled between August 2023 and August 2024. They included 44 with HER2-zero,65 with HER2-low,and 41 with HER2-positive. All underwent ultrasonographic viscoelastic breast imaging preoperatively,wherein the viscoelastic parameters of the target lesion and surrounding 2 mm were measured. Data pre-processing was completed before modeling,and three predictive models of HER2 expression in breast cancer were constructed,based on five machine-learning algorithms. An explanatory analysis of the optimal machine learning model was performed using the SHAP method. Results The triple-classification machine-learning models based on ultrasonographic viscoelastic imaging technology showed favorable performance in terms of predicting the state of HER2 expression in breast cancer. In the verification cohort,the Light Gradient Boosting Machine (LightGBM) model predicted HER2 macro-average AUC,accuracy,precision,recall,and F1-scores of 0.827,0.875,0.860,0.863,and 0.860,respectively. The SHAP analysis results showed that the maximum intratumor elasticity value (Emax) of breast cancer lesions in the LightGBM model contributed the most to the prediction of HER2-zero and HER2-positive,and the maximum peritumoral-2 mm dispersion coefficient (Dmax_2 mm) of breast cancer lesions contributed the most to the prediction of HER2-low. Conclusion This triple-classification machine learning model based on ultrasonographic viscoelastic imaging technology is capable of non-invasively predicting HER2 expression in patients with breast cancer prior to surgical treatment. An interpretable analysis of the model is achievable using the SHAP method.

关键词

超声黏弹性成像技术 / 乳腺癌 / 人表皮生长因子受体2 / 机器学习 / 沙普利附加解释

Key words

ultrasonic viscoelastic imaging technology / breast cancer / human epidermal growth factor receptor 2 / machine learning / Shapley additive explanations

引用本文

引用格式 ▾
覃坚钊,黄群,袁晗,覃秋群,廖新红,高泳. 基于超声黏弹性成像技术构建预测乳腺癌HER2三分类表达的可解释性机器学习模型[J]. 中国医科大学学报, 2026, 55(6): 502-506 DOI:10.12007/j.issn.0258‐4646.2026.06.004

登录浏览全文

4963

注册一个新账户 忘记密码

参考文献

[1]

BRAY F,LAVERSANNE M,SUNG H,et al. Global cancer statistics 2022:GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries [J]. CA Cancer J Clin,2024,74 (3):229-263. DOI:10.3322/caac.21834.

[2]

中国临床肿瘤学会乳腺癌专家委员会,中国抗癌协会乳腺癌专业委员会. 人表皮生长因子受体2阳性乳腺癌临床诊疗专家共识 (2021版) [J]. 中华医学杂志,2021,101 (17):1226-1231. DOI:10.3760/cma.j.cn112137-20210318-00679.

[3]

白冰,张艾佳,郭新,等. HER2低表达乳腺癌的临床病理特征和预后因素分析 [J]. 中国医科大学学报,2023,52 (6):494-498. DOI:10.12007/j.issn.0258-4646.2023.06.003.

[4]

EIGER D,AGOSTINETTO E,SAÚDE-CONDE R,et al. The exciting new field of HER2-low breast cancer treatment [J]. Cancers,2021,13 (5):1015. DOI:10.3390/cancers13051015.

[5]

聂佳慈,勾文峰,许飞飞,等. 缺氧条件下外泌体促进乳腺癌发展的研究进展 [J]. 沈阳药科大学学报,2025,42 (10):896-903.DOI:10.14066/j.cnki.cn21-1349/r.2024.0665.

[6]

JIA WR,XIA SJ,JIA XH,et al. Ultrasound viscosity imaging in breast lesions:a multicenter prospective study [J]. Acad Radiol,2024,31 (9):3499-3510. DOI:10.1016/j.acra.2024.03.017.

[7]

黄志彬,唐淑珍,丁志敏,等. 慢性乙型肝炎患者超声黏弹性指标与临床血清学指标的相关性研究 [J]. 影像研究与医学应用,2023,7 (18):72-75. DOI:10.3969/j.issn.2096-3807.2023.18.024.

[8]

ZHANG XQ,ZHENG RQ,JIN JY,et al. US shear-wave elastography dispersion for characterization of chronic liver disease [J]. Radiology,2022,305 (3):597-605. DOI:10.1148/radiol.212609.

[9]

王坤,朱宇莉,陈凯玲,等. 剪切波频散成像鉴别诊断肝肿瘤良恶性的初步应用 [J]. 中华超声影像学杂志,2022,31 (6):518-524. DOI:10.3760/cma.j.cn131148-20211115-00834.

[10]

张琪,董怡,杨道辉,等. 超声剪切波黏弹性技术在肝肿瘤良恶性鉴别中的初步临床应用 [J]. 中华超声影像学杂志,2019,28 (9):766-770. DOI:10.3760/cma.j.issn.1004‐4477.2019.09.006.

[11]

舒予静,何子龙,曾伟雄,等. 基于双模态乳腺影像特征的HER2表达状态多分类机器学习预测模型构建及可解释性分析研究 [J]. 临床放射学杂志,2024,43 (8):1310-1316. DOI:10.13437/j.cnki.jcr.2024.08.016.

[12]

ŁUKASIEWICZ S,CZECZELEWSKI M,FORMA A,et al. Breast cancer-epidemiology,risk factors,classification,prognostic markers,and current treatment strategies-an updated review [J]. Cancers,2021,13 (17):4287. DOI:10.3390/cancers13174287.

[13]

李艳,栾宇婷,常静瑄,等. 剪切波弹性成像参数对乳腺肿块的鉴别价值及与临床病理学特征的相关性分析 [J]. 肿瘤影像学,2023,32 (6):506-511. DOI:10.19732/j.cnki.2096-6210.2023.06.004.

[14]

JIN J,LI B,CAO JN,et al. Analysis of clinical features,genomic landscapes and survival outcomes in HER2-low breast cancer [J]. J Transl Med,2023,21 (1):360. DOI:10.1186/s12967-023-04076-9.

[15]

BAE MS,KIM HY,OH H,et al. Clinical applications of shear wave dispersion imaging for breast lesions:a pictorial essay [J]. Ultrasonography,2023,42 (4):589-599. DOI:10.14366/usg.23079.

AI Summary AI Mindmap
PDF (1241KB)

0

访问

0

被引

详细

导航
相关文章

AI思维导图

/

〈 〉