Breast cancer molecular subtyping (luminal A, luminal B, HER2-enriched, and triple-negative) guides systemic therapy selection and prognostication, yet is still determined primarily by invasive tissue sampling. Over the last decade, ultrasound has progressed from descriptive B-mode signs to quantitative vascular, mechanical, and computational phenotyping intended to support preoperative subtype inference and biomarker-related risk stratification. This review synthesizes recent advances in conventional ultrasound feature analysis, elastography, contrast-enhanced ultrasound (CEUS), microvascular imaging, radiomics, and deep learning, with emphasis on methodological rigor, interpretability, and clinical translation. Across cohorts, vascularity-related signals (CEUS perfusion metrics and superb microvascular imaging vascular index), stiffness measurements from shear-wave elastography, and multiparametric machine-learning models repeatedly show subtype-discriminative potential. However, heterogeneity in reference standards, single-center retrospective designs, class imbalance, and limited external validation remain major barriers. We outline a roadmap toward clinically deployable ultrasound-based subtyping—prioritizing standardized acquisition, prospective multicenter evaluation, uncertainty-aware and interpretable AI, and multimodal integration with clinicopathologic and genomic context. Importantly, subtype separability should be interpreted as probabilistic discrimination of operational labels rather than biological determinism; mechanistic origins remain largely unknown and require dedicated radiologic–pathologic/radiogenomic validation.
Hong et al. (Wed,) studied this question.