Please wait a minute...
图/表 详细信息
MRI预测乳腺癌淋巴结状态的研究进展及展望
翟梓涵, 陈盛
中国癌症杂志    2025, 35 (8): 799-807.   DOI: 10.19401/j.cnki.1007-3639.2025.08.009
摘要   (42 HTML2 PDF(pc) (945KB)(21)  

乳腺癌是全球范围内女性常见的癌症类型及死亡原因,淋巴结状态是乳腺癌分期的重要信息,与患者预后密切相关。磁共振成像(magnetic resonance imaging,MRI)技术对淋巴结以及新辅助治疗应答效果的评估具有优势,可补充其他影像学检查的不足。标准化评分系统如淋巴结数据和报告系统(Node Reporting and Data System,Node-RADS)通过整合淋巴结大小、边缘、强化模式等特征,可有效地减少评估的主观差异。影像组学通过高通量提取定量特征,将医学图像转换为可挖掘的数据并对其进行分析,进一步整合MRI影像组学、临床病理学特征及分子亚型信息构建多组学模型,可有效地预测腋窝淋巴结转移,为个性化治疗提供生物学依据。人工智能能够通过广泛搜索模型和参数空间来生成预测模型,人工智能驱动的MRI影像分析可有效地预测淋巴结转移及治疗反应。在新辅助化疗评估中,基于深度学习的全自动集成系统(fully automated-integrated system based on deep learning,FAIS-DL)结合多区域动态对比增强-MRI(dynamic contrast enhanced-MRI,DCE-MRI)和临床数据可高效能地预测腋窝病理学完全缓解,将不必要的腋窝淋巴结清扫术率从47.9%降至6.8%。本文就不同发展阶段采用MRI预测乳腺癌淋巴结状态的研究进展作一综述,以期提高临床医师和影像科医师对MRI在乳腺癌淋巴结状态评估及新辅助治疗效果评价中应用的认知,并为精准预测乳腺癌淋巴结状态模型的构建提供帮助。


Category Research advances References
Conventional MRI DCE Established the standardized scoring system Node-RADS, providing a quantitative basis for biopsy decision-making [9-10]
MRL Sensitive to the morphology of lymphatic vessels and lymph nodes.
Suitable for internal mammary lymph node examination
[14-16]
MRS When the total choline level is <2.4 mmol/L, no metastatic lymph nodes are detected.
When it is >15 mmol/L, the probability of lymph node metastasis is 2.69 times that in the situation of total choline level <15 mmol/L
[18-19]
DWI In clinical scenarios of contrast agent allergy or renal insufficiency, morphological assessment of axillary lymph nodes on DWI can distinguish benign from malignant, and clearly visualize internal mammary lymph nodes [22-26]
MRI radiomics Combined with other features Radiomics models integrating clinical, pathological, intra-tumoral, and peritumoral (3 or 4 mm) features demonstrate favorable predictive performance [54-55]
Based on machine learning The XGBoost algorithm exhibits relatively stable performance.
Multi-spatiotemporal model assisting surgical strategies can effectively reduce the false-negative rate of sentinel lymph nodes
[58-60,63]
Based on deep learning The CNN model demonstrates good predictive performance.
The fully automatic integrated system FAIS-DL, which is based on deep learning, can efficiently predict pathological complete response of axillary lymph nodes, thereby helping to reduce unnecessary axillary lymph node dissection
[65-70,75 -76]
View table in article
表1 MRI预测乳腺癌淋巴结状态的主要研究进展
正文中引用本图/表的段落
Li等[75]提出了一种基于深度学习的全自动集成系统(fully automated-integrated system based on deep learning,FAIS-DL),利用临床病理学特征、肿瘤和ALN的DCE-MRI预测乳腺癌NAT后的腋窝pCR。FAIS-DL在内部测试集、合并外部测试集和前瞻性测试集中AUC分别达到了0.95、0.93和0.94,显著高于基于单区域DCE-MRI的临床模型和深度学习模型。在合并的外部和前瞻性测试集中,FAIS-DL将ALND率从47.9%降低到6.8%,并将受益率从52.2%提高到86.5%,有较好的临床适用性。有研究[76]指出,DL模型可以提供一种非手术方法通过乳腺MRI自动预测ALN转移,多参数MRI和结合多个模型的集成学习值得进一步研究。MRI预测乳腺癌淋巴结状态的主要研究进展见表1。
DCE: Dynamic contrast enhanced; Node-RADS: Node Reporting and Data System; MRL: Magnetic resonance lymphography; MRS: Magnetic resonance spectroscopic imaging; DWI: Diffusion weighted imaging; XGBoost: Extreme gradient boosting; CNN: Convolutional neural network. ...
Deep learning radiomics of preoperative breast MRI for prediction of axillary lymph node metastasis in breast cancer
1
2023
... Chen等[71]研究了479例乳腺癌患者共计488个病灶的术前磁共振成像数据,开发了应用DenseNet 121的预训练神经网络从DCE-MRI和DWI-ADC中提取影像组学特征,进一步将影像组学特征与淋巴结可触及性、MRI中的肿瘤大小和Ki-67增殖指数等临床病理学特征相结合,训练集的AUC从0.76提高到了0.80. ...
Attention-based deep learning for the preoperative differentiation of axillary lymph node metastasis in breast cancer on DCE-MRI
1
2023
... 一项回顾性研究[72]纳入941例术前接受DCE的乳腺癌患者,提出了一种基于3D深度ResNet架构和卷积块注意力模块的深度学习模型(RCNet)用于ALN转移识别,该模型在内部测试集中AUC为0.907,外部验证集中AUC为0.853.RCNet模型的灰度图像和相应的热图显示原发肿瘤和ALN这两个区域对识别淋巴结状态很有价值. ...
Multimodal deep learning fusion of ultrafast-DCE MRI and clinical information for breast lesion classification
1
2025
... Lokaj等[73]融合超快动态对比增强MRI(ultrafast dynamic contrast-enhanced MRI,UF-DCE MRI)图像、病变特征和临床信息,对比了传统机器学习方法、单模态影像模型及多模态编码器算法MMST-V的性能.结果显示,其MMST-V性能显著优于仅基于临床信息或影像数据的模型,表明多模态深度学习融合影像与临床信息可提升乳腺病变分类的准确性. ...
Development of MRI-based deep learning signature for prediction of axillary response after NAC in breast cancer
1
2024
... 为研究基于深度学习的MRI影像组学在NAT后乳腺癌ALN状态中的效用,Zhang等[74]的回顾性研究纳入了327例患者,在NAT前的DCE图像上识别原发肿瘤并进行三维分割,使用ResNet34提取深度学习特征预测NAT后淋巴结状态,其中支持向量机模型在训练集和测试集中的AUC分别为0.99和0.83,展现出最好的性能.放射组学特征和临床特征整合建立的列线图AUC为0.99. ...
Multiregional dynamic contrast-enhanced MRI-based integrated system for predicting pathological complete response of axillary lymph node to neoadjuvant chemotherapy in breast cancer: multicentre study
2
2024
... Li等[75]提出了一种基于深度学习的全自动集成系统(fully automated-integrated system based on deep learning,FAIS-DL),利用临床病理学特征、肿瘤和ALN的DCE-MRI预测乳腺癌NAT后的腋窝pCR.FAIS-DL在内部测试集、合并外部测试集和前瞻性测试集中AUC分别达到了0.95、0.93和0.94,显著高于基于单区域DCE-MRI的临床模型和深度学习模型.在合并的外部和前瞻性测试集中,FAIS-DL将ALND率从47.9%降低到6.8%,并将受益率从52.2%提高到86.5%,有较好的临床适用性.有研究[76]指出,DL模型可以提供一种非手术方法通过乳腺MRI自动预测ALN转移,多参数MRI和结合多个模型的集成学习值得进一步研究.MRI预测乳腺癌淋巴结状态的主要研究进展见表1. ...

DCE: Dynamic contrast enhanced; Node-RADS: Node Reporting and Data System; MRL: Magnetic resonance lymphography; MRS: Magnetic resonance spectroscopic imaging; DWI: Diffusion weighted imaging; XGBoost: Extreme gradient boosting; CNN: Convolutional neural network. ...
Clinical applications of deep learning in breast MRI
2
2023
... Li等[75]提出了一种基于深度学习的全自动集成系统(fully automated-integrated system based on deep learning,FAIS-DL),利用临床病理学特征、肿瘤和ALN的DCE-MRI预测乳腺癌NAT后的腋窝pCR.FAIS-DL在内部测试集、合并外部测试集和前瞻性测试集中AUC分别达到了0.95、0.93和0.94,显著高于基于单区域DCE-MRI的临床模型和深度学习模型.在合并的外部和前瞻性测试集中,FAIS-DL将ALND率从47.9%降低到6.8%,并将受益率从52.2%提高到86.5%,有较好的临床适用性.有研究[76]指出,DL模型可以提供一种非手术方法通过乳腺MRI自动预测ALN转移,多参数MRI和结合多个模型的集成学习值得进一步研究.MRI预测乳腺癌淋巴结状态的主要研究进展见表1. ...

本文的其它图/表