Ultrasound Elastography Radiomics: A Novel Approach for Benign-Malignant Differentiation of BI-RADS Category 4 Breast...

To develop a non-invasive, accurate diagnostic model by integrating ultrasound radiomics with Ultrasound Elastography (UE) for BI-RADS category 4 breast masses, addressing limitations of current diagnostic methods.

A total of 720 patients with BI-RADS category 4 breast lesions (504 for training/216 for internal validation) and 102 external validation patients were enrolled. From the 623 radiomic features extracted by Pyradiomics, 9 stable features were retained after Z-score standardization, intraclass correlation coefficient analysis (>0.9), and max-relevance and min-redundancy filtering. Three Logistic Regression models (ultrasound radiomics, UE radiomics, and the combined model) were developed and validated using 10-fold cross-validation.

In the internal validation cohort, the combined model (AUC = 0.926) achieved superior diagnostic performance compared to the ultrasound radiomics model (AUC = 0.726, Z = 5.643, P < 0.01), UE radiomics model (AUC = 0.853, Z = 3.021, P = 0.003), and the BI-RADS category (AUC = 0.869, Z = 2.051, P = 0.040). Similar results were observed in the external validation cohort. The combined model demonstrated favorable calibration performance and clinical value, as evidenced by calibration curves and decision curve analysis. The combined radiomic model exhibited high diagnostic specificity in the low-grade malignant cohort (4A, 4B).

For BI-RADS category 4 breast masses, the combined radiomic model enables accurate and non-invasive differentiation of benign and malignant lesions, reduces unnecessary core needle biopsies, and provides a valuable tool for personalized clinical decision-making and breast cancer management.