Radiomics and Deep Learning: Bridging Breast Cancer Imaging Phenotypes and Genomic Heterogeneity

Breast cancer is a highly heterogeneous malignancy. Radiomics, combined with deep learning techniques, harnesses the power of advanced medical imaging to extract rich quantitative features that noninvasively capture tumor imaging phenotypes. This approach facilitates the integration of macroscopic imaging data with genomic heterogeneity, thereby bridging the gap between tumor radiographic appearance and underlying molecular profiles. This review systematically outlines the radiomics workflow in breast cancer, encompassing image acquisition, tumor segmentation, feature extraction, feature selection, and modeling strategies. Emphasis is placed on the role of deep learning in automating feature extraction and enabling multimodal data fusion to enhance predictive accuracy. By examining current research progress, we elucidate methods to uncover latent information related to gene mutations, immune microenvironment characteristics, and other genomic alterations from imaging data. This synthesis highlights the potential of radiogenomics to provide novel insights and tools for precision medicine, ultimately fostering personalized diagnosis, prognosis, and treatment planning in breast cancer care.