A preoperative ultrasound-based radiomics-clinical nomogram to assess metastasis of ipsilateral supraclavicular lymph nodes in patients with primary breast cancer: A biomarker based on breast and axillary images.

BACKGROUND

Breast cancer patients with ipsilateral supraclavicular lymph node (ISLN) metastasis often experience poor prognosis. A non-invasive method for preoperative diagnosis of ISLN metastasis is warranted.

METHODS

This study included breast cancer patients with suspected ISLN involvement. After randomly dividing patients into training and validation groups at 3:2 ratio, radiomic models were constructed using ultrasound (US) images of the breast, axilla, and ISLN. A radiomic score (radscore) was developed using Student's t-test, Fisher's correlation coefficient, and LASSO regression, with 5-fold cross-validation (CV) applied during model construction. Model performance was evaluated using Receiver Operating Characteristic (ROC) curves in the validation group. A nomogram combining the radscore (from the optimal radiomic model) and clinicopathological variables was developed in the training group.

RESULTS

The study involved 547 ultrasound images from 186 eligible patients (112 in the training group and 74 in the validation group). The model combining radiomic features of breast and axillary ultrasound images based on LASSO regression demonstrated superior performance. A nomogram incorporating clinicopathological characteristics (including T-stage, HER-2 status and CA15-3) and radscore (comprising 10 radiomic features) was constructed. The area under the ROC curve (AUC) for ISLN diagnosis were 0.876 (95% CI 0.818-0.941, P < 0.001) and 0.837 (95% CI 0.733-0.952, P = 0.026) in the CV and validation groups, respectively.

CONCLUSION

Radiomic features derived from breast and axillary US images are valuable for assessing ISLN metastasis in breast cancer patients. The preoperative US-based radiomic-clinical nomogram effectively predicts ISLN metastasis non-invasively.