Ultrasound and MRI radiogenomics in breast cancer: a narrative synthesis study of clinical applications and methodological limitations

Breast cancer remains the most prevalent malignancy among women globally, with clinical outcomes driven by molecular subtype heterogeneity, tumour microenvironment dynamics, and variable treatment responses. Radiogenomics—the integration of quantitative imaging features (radiomics) with genomic and multi-omic data—has attracted considerable research interest as a potential approach to non-invasive tumour characterisation. This article is a narrative review of the literature on ultrasound and MRI radiomics, genomic profiling, and their integration in breast cancer, framed around diagnostic, prognostic, treatment-response, and molecular-subtype applications. We aim to provide a balanced synthesis that distinguishes demonstrated clinical utility from preliminary, hypothesis-generating findings. We searched PubMed/MEDLINE, EMBASE, Web of Science, and the Cochrane Library for representative English-language publications from January 2017 to April 2025, supplemented by landmark Phase II/III trials, foundational genomic profiling studies, and population-based epidemiological analyses. No formal systematic-review methodology (PRISMA), formal quality-appraisal instrument (RQS, AMSTAR-2, QUADAS-2, PROBAST, CLEAR), or formal certainty-rating framework (GRADE) was applied; appraisal of the evidence is narrative and interpretive. A representative catalogue of studies discussed in this review is provided in Table 6. Across the representative evidence base, DCE-MRI radiomics has achieved AUCs of approximately 0.84–0.95 for benign–malignant differentiation, molecular subtype classification, and neoadjuvant chemotherapy (NAC) response prediction, and ultrasound-based radiomics has achieved AUCs of approximately 0.84–0.94 for comparable diagnostic tasks, with performance varying by clinical task, study population, reference standard, and validation design. Selected integrated radiogenomic models have shown numerically higher discrimination than single-modality comparators in some studies, but the gain is not consistent across the literature and has not been prospectively validated for treatment selection. Established targeted therapies guided by molecular profiling have demonstrated significant survival benefits in randomised trials; however, no trial has yet shown that radiogenomic-guided treatment decisions improve patient outcomes. Current evidence suggests that radiogenomics is a scientifically compelling but methodologically immature field. Most published models are retrospective, single-centre, and lack external multi-institutional validation; reported AUCs should be interpreted as upper bounds rather than expected real-world performance. Radiogenomics cannot at present substitute for tissue-based molecular profiling. Standardisation of radiomic workflows, adequately powered prospective trials with clearly specified intended uses, and formal clinical-utility evaluation are prerequisites for clinical translation.