Diagnostic performance of multimodal ultrasonography for molecular subtyping of breast cancer: a retrospective study

Breast cancer molecular subtypes differ substantially in treatment selection, therapeutic response, and prognosis, making accurate subtype identification essential for individualized management. Current molecular subtyping mainly relies on immunohistochemical and molecular pathological analyses of biopsy or surgical specimens, which are invasive and may be affected by sampling bias, intratumoral heterogeneity, and testing turnaround time. This study aimed to evaluate the diagnostic performance of multimodal ultrasonography for noninvasive molecular subtyping of breast cancer.

A total of 155 patients diagnosed with breast cancer were retrospectively enrolled. According to molecular classification, patients were categorized into the human epidermal growth factor receptor 2 (HER2)-overexpression subtype (n=32), triple-negative subtype (n=31), luminal A subtype (n=26), and luminal B subtype (n=66). Differences in multimodal ultrasound features among the various molecular subtypes were compared, and the predictive value of multimodal ultrasound parameters for molecular classification was analyzed.

Higher maximum elastic modulus (E_max) and lesion and elasticity ratio between the lesion and surrounding normal tissue (E_ratio) values were observed in the HER2-overexpression and triple-negative subtypes than in the luminal A and luminal B subtypes (P<0.05). Multivariate logistic regression analysis identified calcification status, Adler blood flow grading, vascularization index (VI), and E_ratio as independent predictors of breast cancer molecular subtype (P<0.05). Receiver operating characteristic curve analysis showed that the combined multimodal ultrasonographic model demonstrated favorable diagnostic performance in identifying different molecular subtypes of breast cancer, with area under the curve (AUC) of 0.856 for HER2-overexpression, 0.973 for triple-negative, 0.900 for luminal A, and 0.811 for luminal B subtypes. The corresponding sensitivity/specificity values were 93.8%/71.5%, 96.8%/87.9%, 88.5%/82.2%, and 89.4%/66.3%, respectively. DeLong's test showed that the AUCs of the combined model were significantly higher than those of the individual ultrasound parameters for all four molecular subtypes (all P<0.05).

The molecular subtypes of breast cancer exhibited distinct multimodal ultrasonographic characteristics. Multimodal ultrasonography showed potential value for noninvasive prediction of breast cancer molecular subtypes and may provide auxiliary information for individualized therapeutic decision-making.