Completion axillary lymph node dissection is overtreatment for patients with sentinel lymph node (SLN) metastasis in whom the metastatic risk of residual non-SLN (NSLN) is low. However, the National Comprehensive Cancer Network panel posits that none of the previous studies has successfully identified such subset patients. Here, we develop a multicentre deep learning radiomics of ultrasonography model (DLRU) to predict the risk of SLN and NSLN metastasis. In total, 937 eligible breast cancer patients with ultrasound images were enrolled from two hospitals as the training set (n = 542) and independent test set (n = 395) respectively. Using the images, we developed and validated a prediction model combined with deep learning radiomics and axillary ultrasound to sequentially identify the metastatic risk of SLN and NSLN, thereby, classifying patients to relevant axillary management groups. In the test set, the DLRU yields the best performance in identifying patients with metastatic disease in SLNs (sensitivity=98.4%, 95% CI 96.6–100) and NSLNs (sensitivity=98.4%, 95% CI 95.6–99.9). The DLRU also accurately stratifies patients without metastasis in SLN or NSLN into the corresponding low-risk (LR)-SLN and high-risk (HR)-SLN&LR-NSLN category with the negative predictive value of 97% (95% CI 94.2–100) and 91.7% (95% CI 88.8–97.9), respectively. Moreover, compared with the current clinical management, DLRU appropriately assigned 51% (39.6%/77.4%) of overtreated patients in the entire study cohort into the LR group, perhaps avoiding overtreatment. The performance of the DLRU indicates that it may offer a simple preoperative tool to promote personalized axillary management of breast cancer. The National Nature Science Foundation of China; The National Outstanding Youth Science Fund Project of National Natural Science Foundation of China; The Scientific research project of Heilongjiang Health Committee; The Postgraduate Research &Practice Innovation Program of Harbin Medical University.

Breast cancer is molecularly heterogeneous and categorized into four molecular subtypes: Luminal-A, Luminal-B, HER2-amplified and Triple-negative. In this study, we aimed to apply an ensemble decision approach to identify the ultrasound and clinical features related to the molecular subtypes. We collected ultrasound and clinical features from 1,000 breast cancer patients and performed immunohistochemistry on these samples. We used the ensemble decision approach to select unique features and to construct decision models. The decision model for Luminal-A subtype was constructed based on the presence of an echogenic halo and post-acoustic shadowing or indifference. The decision model for Luminal-B subtype was constructed based on the absence of an echogenic halo and vascularity. The decision model for HER2-amplified subtype was constructed based on the presence of post-acoustic enhancement, calcification, vascularity and advanced age. The model for Triple-negative subtype followed two rules. One was based on irregular shape, lobulate margin contour, the absence of calcification and hypovascularity, whereas the other was based on oval shape, hypovascularity and micro-lobulate margin contour. The accuracies of the models were 83.8%, 77.4%, 87.9% and 92.7%, respectively. We identified specific features of each molecular subtype and expanded the scope of ultrasound for making diagnoses using these decision models.