Abstract
OBJECTIVE: To evaluate the diagnostic performance of mammography and ultrasonography in distinguishing benign from malignant breast structural distortions and to develop an integrated predictive model combining radiomic features and molecular markers for improved risk stratification. METHODS: This retrospective study included 260 patients with histopathologically confirmed breast structural distortions (156 malignant, 104 benign). Lesions were characterized using Breast Imaging Reporting and Data System (BI-RADS) criteria. Radiomic features were extracted with PyRadiomics, harmonized via ComBat, and selected using Least Absolute Shrinkage and Selection Operator (LASSO) regression. A predictive model incorporated imaging features, molecular markers (vascular endothelial growth factor [VEGF], transforming growth factor-β1 [TGF-β1]), and clinical variables. Diagnostic accuracy was assessed by sensitivity, specificity, AUC, and decision curve analysis, with subgroup analyses by age, menopausal status, and breast density. RESULTS: Malignant distortions showed higher rates of spiculated margins (82.1% vs. 16.3%, P<0.001) and hypoechoic irregular masses (78.2% vs. 27.9%, P<0.001). Combined mammography-ultrasound assessment improved diagnostic performance (AUC 0.91) versus single modalities (mammography 0.79; ultrasound 0.82). The radiomic-molecular model further enhanced accuracy (AUC 0.94) and reduced unnecessary biopsies by 32%. Spiculation complexity and VEGF overexpression were independent predictors of lymphovascular invasion and lower 5-year disease-free survival (68% vs. 89%, P=0.01). CONCLUSION: Integrating mammography, ultrasonography, and radiomic-pathologic markers significantly improves differentiation of malignant breast distortions and supports personalized prognosis.