Related Experiment Video
Updated: Mar 9, 2026
10:02
High-frequency Ultrasound Imaging of Mouse Cervical Lymph Nodes
Published on: July 25, 2015
19.8K
Ultrasound Video-Based Deep Learning Model for Predicting Axillary Lymph Node Status and Nodal Burden in Breast
Jiaheng Huang1, Qing Xia2, Yuqi Yan3
1Department of Diagnostic Ultrasound Imaging & Interventional Therapy, Zhejiang Cancer Hospital, Hangzhou, Zhejiang, China (J.H., Y.Y., Z.C., Q.L., Y.Z., C.C., X.Z.); Wenzhou Medical University, Wenzhou, Zhejiang 325035, China (J.H., Z.J., X.Z., D.X.).
Academic Radiology
|March 7, 2026
Summary
A novel deep learning (DL) framework using breast ultrasound videos accurately predicts axillary lymph node (ALN) status in breast cancer patients. The model also shows moderate performance in predicting nodal burden, aiding personalized treatment.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Oncology
- Breast Cancer Diagnostics
Background:
- Accurate preoperative assessment of axillary lymph node (ALN) status and nodal burden is critical for tailoring breast cancer management.
- Current methods may have limitations in precisely determining ALN involvement before surgery.
Purpose of the Study:
- To develop and validate a two-stage deep learning (DL) framework utilizing preoperative breast ultrasound videos.
- To predict both ALN status (negative vs. positive) and nodal burden (1-2 vs. ≥3 nodes) in breast cancer patients.
Main Methods:
- A multicenter retrospective study involving 864 breast cancer patients.
- A two-stage DL framework based on a Temporal Shift Module (TSM) video model was employed.
- Model performance was assessed using AUC, sensitivity, and specificity across training, internal, and external test sets.
Main Results:
- The TSM-ResNet50 model achieved high AUCs for ALN status prediction (0.851-0.886) across test sets, with consistent performance in subgroups.
- The TSM-ResNet18 model demonstrated promising AUCs for nodal burden prediction (0.667-0.937), indicating clinical relevance.
- The DL framework showed robust performance, particularly for ALN status prediction.
Conclusions:
- A two-stage video-based DL model shows significant potential for preoperative axillary assessment in breast cancer.
- The framework offers promising capabilities for predicting ALN status and clinically meaningful performance for nodal burden.
- This approach may enhance individualized patient management by providing crucial preoperative insights.
