Deep learning (DL) algorithms have been proven to be very effective in a wide range of computer vision applications, such as segmentation, classification, and detection. DL models can automatically assess complex medical image scenes without human intervention and can be applied as a second reader to provide an additional opinion for the physician. To predict the axillary lymph node (ALN) metastatic status in patients with early-stage breast cancer, a deep learning-based computer-aided prediction system for ultrasound (US) images was proposed. A total of 153 women with breast tumor US images were involved in this study; there were 59 patients with metastasis and 94 patients without ALN metastasis. A deep learning-based computer-aided prediction (CAP) system using the tumor region and peritumoral tissue in ultrasound (US) images were employed to determine the ALN status in breast cancer. First, we adopted Mask R–CNN as our tumor detection and segmentation model to obtain the tumor localization and region. Second, the peritumoral tissue was extracted from the US image, which reflects metastatic progression. Third, we used the DL model to predict ALN metastasis. Finally, the simple linear iterative clustering (SLIC) superpixel segmentation method and the LIME explanation algorithm were employed to explain how the model makes decisions. The experimental results indicated that the DL model had the best prediction performance on tumor regions with 3 mm thick peritumoral tissue, and the accuracy, sensitivity, specificity, and AUC were 81.05% (124/153), 81.36% (48/59), 80.85% (76/94), and 0.8054, respectively. The results indicated that the proposed CAP system could help determine the ALN status in patients with early-stage breast cancer. The results reveal that the proposed CAP model, which combines primary tumor and peritumoral tissue, is an effective method to predict the ALN status in patients with early-stage breast cancer.

深度学习（DL）算法已被证明在广泛的计算机视觉应用中非常有效，例如分段，分类和检测。DL模型可以自动评估复杂的医学图像场景，而无需人工干预，并且可以用作第二阅读器，为医生提供其他意见。为了预测早期乳腺癌患者的腋窝淋巴结（ALN）转移状态，提出了一种基于深度学习的计算机辅助超声（US）图像预测系统。共有153名具有乳腺US图像的女性参与了这项研究。有转移的59例和没有ALN转移的94例。使用基于深度学习的计算机辅助预测（CAP）系统，该系统使用超声（US）图像中的肿瘤区域和肿瘤周围组织来确定乳腺癌中ALN的状态。首先，我们采用口罩[R-CNN作为我们的肿瘤检测和分割模型，以获得肿瘤的定位和区域。第二，从US图像中提取肿瘤周围组织，其反映了转移进展。第三，我们使用DL模型来预测ALN转移。最后，采用简单的线性迭代聚类（SLIC）超像素分割方法和LIME解释算法来解释模型如何做出决策。实验结果表明，DL模型在肿瘤周围组织厚度为3 mm的肿瘤区域具有最佳的预测性能，准确度，灵敏度，特异性和AUC分别为81.05％（124/153），81.36％（48/59），分别为80.85％（76/94）和0.8054。结果表明，提出的CAP系统可以帮助确定早期乳腺癌患者的ALN状况。