Machine learning-driven ultrasound radiomics for assessing axillary lymph node burden in breast cancer

• Published in: Frontiers in endocrinology (Lausanne) Vol. 16; p. 1548888
• Main Authors: Wang, Si-Rui, Tian, Feng, Zhu, Tong, Cao, Chun-Li, Wang, Jin-Li, Li, Wen-Xiao, Li, Jun, Hou, Ji-Xue
• Format: Journal Article
• Language: English
• Published: Switzerland Frontiers Media S.A 27.02.2025
• Subjects: Adult; Aged; Axilla; axillary lymph nodes burden; breast cancer; Breast Neoplasms - diagnostic imaging; Breast Neoplasms - pathology; Endocrinology; Female; Humans; Lymph Nodes - diagnostic imaging; Lymph Nodes - pathology; Lymphatic Metastasis - diagnostic imaging; Machine Learning; Middle Aged; Nomograms; Radiomics; Retrospective Studies; Ultrasonography - methods; ultrasound; radiomics; axillary lymph nodes burden; breast cancer; machine learning; ultrasound
• DOI: 10.3389/fendo.2025.1548888
• ISSN: 1664-2392

This study explores the value of combining intratumoral and peritumoral radiomics features from ultrasound imaging with clinical characteristics to assess axillary lymph node burden in breast cancer patients. A total of 131 breast cancer patients with axillary lymph node metastasis (ALNM) were enrolled between June 2019 and September 2024. Patients were divided into low (n=79) and high (n=52) axillary lymph node burden (ALNB) groups. They were further split into training (n=92) and validation (n=39) cohorts. Intratumoral and peritumoral features were analyzed using the maximum relevance minimum redundancy (MRMR) and least absolute shrinkage and selection operator (LASSO) methods. Six machine learning models were evaluated, and a combined clinical-radiomics model was built. The combined logistic regression model exhibited superior diagnostic performance for high axillary lymph node burden, with areas under the ROC curve (AUC) of 0.857 in the training cohort and 0.820 in the validation cohort, outperforming individual models. The model balanced sensitivity and specificity well at a 52% cutoff value. A nomogram provided a practical risk assessment tool for clinicians. The combined clinical-radiomics model showed excellent predictive ability and may aid in optimizing management and treatment decisions for breast cancer patients.