脉得智能
Recently, two research findings published in internationally authoritative journals have demonstrated the strong application potential of artificial intelligence in the field of ultrasound medicine. MedAI, as a supporter of the core algorithm and model development, was deeply involved in both innovative studies, helping advance clinical diagnosis and treatment into a new stage of precision. Achievement 1: A New Standard for Noninvasive Assessment of Sentinel Lymph Nodes in Breast Cancer A study published in EClinicalMedicine (IF 10) successfully developed a deep learning dual-modal fusion network (DDFN) based on contrast-enhanced ultrasound (CEUS) and grayscale ultrasound (GSUS). The model is designed to noninvasively predict the metastatic status of sentinel lymph nodes (SLNs) in patients with breast cancer before surgery. Its core value lies in its extremely high negative predictive value (NPV) and specificity. The model demonstrated robust performance across multiple prospective multicenter tests, showing excellent prospects for clinical translation. Clinical value: (1) It is expected to enable early-stage breast cancer patients to avoid unnecessary sentinel lymph node biopsy. Its high negative predictive value can accurately identify low-risk patients and avoid invasive procedures. (2) It reduces surgical trauma and complications, optimizes individualized treatment strategies, lowers healthcare resource consumption, and improves patients' quality of life and treatment experience. (3) It promotes the transition of axillary management in breast cancer from pathological staging to imaging staging, providing a feasible pathway for the concept of “imaging N0 (iN0)” and supporting precision surgical decision-making.
△ Flowchart of the development process for the deep learning model used to diagnose sentinel lymph node status.
△ Representative heatmaps generated using the Grad-CAM algorithm for GSUS and CEUS images of SLNs (lymph nodes) with different characteristics.
Achievement 2: An Intelligent Tool for Diagnosis and Risk Stratification of Gallbladder Polyps A study published in Insights into Imaging (IF 4.5) developed an integrated deep learning (IDL) model based on grayscale ultrasound and color Doppler flow imaging (CDFI). This model can not only automatically and accurately segment the gallbladder, but also effectively differentiate non-neoplastic from neoplastic polyps, as well as benign from malignant polyps. It delivers excellent diagnostic performance. More importantly, as an assistive tool, it can significantly improve the diagnostic capability and consistency of less-experienced ultrasound physicians. It is expected to reduce the risks of overtreatment or missed diagnosis arising from decisions based on a single size criterion (such as 10 mm), providing an objective basis for individualized and precise management of gallbladder polyps.
△ Diagnostic performance of each deep learning model on the validation set.
△ Workflow of the IDL model
MedAI's Contributions and Clinical Significance In the above groundbreaking studies, the MedAI team provided critical support in algorithm design, model development, and optimization. Using advanced deep learning architectures, including dual-modal fusion and cross-attention mechanism modules, we achieved deep feature extraction and effective fusion of dual-modal ultrasound images. This was central to the models' high accuracy and strong generalizability. The clinical implications of these two achievements are far-reaching: by integrating artificial intelligence with routine ultrasound examinations, they provide objective, reproducible, and efficient AI-assisted decision-making tools for addressing clinical challenges such as precision axillary management in breast cancer and risk stratification of gallbladder polyps. This not only helps advance precision medicine, optimize treatment strategies, and reduce unnecessary invasive procedures and surgical complications, but also improves diagnostic capabilities in primary-level hospitals, benefiting a broader patient population. MedAI will continue to pursue the deep integration of cutting-edge artificial intelligence technologies with clinical needs, empowering medical imaging and supporting precision medicine. References:
- Huang, D., Shi, Y., Cao, W., Feng, L., Luo, Y., Zhao, X., ... & Luo, J. (2026). Exempting axillary staging surgery in breast cancer using multimodal ultrasound imaging and radiomics of sentinel lymph nodes. EClinicalMedicine, 92.
- Tang, C., Shi, Y., Wang, L., Zhao, X., Li, C., Guan, P., ... & Yuan, H. (2026). An integrative deep learning model based on dual-mode ultrasound for diagnosing gallbladder polyps. Insights into Imaging, 17(1), 32.
Written and formatted by: Luo Chaoran, Huang Qian Design by: Huang Ruoxi Reviewed by: Li Chunlei, Zhao Xing, Mou Lichao, Chen Yonghong