<p>Breast cancer remains one of the most prevalent and deadly cancers worldwide, with early and accurate detection being critical for improving patient outcomes. However, mammography interpretation is challenged by the complex, low-contrast nature of breast tissue and significant variations across imaging datasets, often leading to diagnostic uncertainty. To address these gaps, we propose HFEN-Breast, a novel Hierarchical Feature Extraction Network specifically designed to enhance breast cancer detection in mammography. HFEN-Breast introduces a multi-stage architecture comprising a Hierarchical Radiomic Extraction Layer (HREL) for robust multi-scale feature capture, a Hierarchical Attention Refinement Block (HARB) to focus on diagnostically significant regions, a Hierarchical Semantic Fusion Block (HSFB) for contextual semantic integration, and a Hierarchical Multi-Scale Fusion Module (HMSFM) for efficient feature consolidation. Evaluated on the CBIS-DDSM and INbreast datasets, HFEN-Breast achieves state-of-the-art performance, with an AUC of 0.987 on INbreast and substantial improvements in classification accuracy and sensitivity over existing methods. These results demonstrate that HFEN-Breast not only addresses the limitations of current deep learning models in handling dataset variability and subtle lesion detection but also offers enhanced reliability and generalizability for clinical breast cancer screening and diagnostics.</p>