前哨淋巴结1～2个阳性的乳腺癌患者中，约2/3其非前哨淋巴结未发生转移，可采用局部放疗代替腋窝淋巴结清扫，但目前尚无准确预测非前哨淋巴转移的方法。申请人和国内外研究团队的研究证实，影像组学及深度学习方法可用于乳腺癌的辅助诊断和评估，尤其在预测腋窝淋巴结转移方面具有良好的前景，但在预测非前哨淋巴结转移中的价值尚不明确。本项目在前期研究基础上，将深度挖掘高时间和空间分辨率DCE-MRI的信息，开展：（1）提取包含动态信息的影像组学特征，融合临床、病理及DCE-MRI参数等信息构建多维度影像组学模型；（2）通过深度学习网络DenseNet-201提取深度特征，输入长短时程记忆（long-short term memory，LSTM）网络构建基于时间序列的深度学习模型；（3）选择最优模型并利用独立样本进行验证。本研究的结果，将为临床医生预测乳腺癌非前哨淋巴结转移提供辅助工具，指导临床决策。

Among the early breast cancer patients with one or two involved sentinel lymph nodes, about 2/3 have no non-sentinel lymph node metastasis, and local radiotherapy can be used instead of axillary lymph node dissection. However, there is no accurate method to predict non-sentinel lymph node metastasis in those patients. Our previous research and other researchers' results indicated that radiomics and deep learning can decode the clinical information including diagnosis and evaluation for breast cancer, especially in the prediction of axillary lymph node metastasis. However, its value in the prediction of non-sentinel lymph node metastasis is not clear. On the basis of previous research, this project will deeply mine the information of DCE-MRI with high temporal and spatial resolution. This project is to: (1) extract radiomics features, and integrate clinical, pathological and DCE-MRI parameters to build a multi-dimensional radiomics model; (2) extract deep features through the deep learning network Densenet-201, and input long-short term memory (LSTM) network constructs deep learning model; (3) select the optimal model and validate it. The results of this project will provide an adjunct tool for clinicians to predict non-sentinel lymph node metastasis of breast cancer and guide clinical decision-making.