This paper presents an all-encompassing process that includes a hybrid deep learning model with GRU, VGG-19 and U-Net architectures for breast cancer detection and classification. This approach starts by selecting and processing mammographic images to standardize input formats and enhance image quality, which is important for accurate segmentation. Hence, the presented hybrid model captures temporal dependencies accurately as well as extracts robust features towards obtaining precise semantic segmentation as shown by the input, preprocessed, and segmented and tumor detected images. Initial results suggest that there is promising improvement in tumor localization and delineation accuracy as these method outperforms established methods such as Random Forest (96.4%), AlexNet (98.60%), U-Net (99.33%) and even Hybrid ABO method (99.6%). It is worth noting that our proposed model has achieved 99.71% accuracy rate which suggests its viability for application in clinical settings. Such a carefully designed framework helps to detect breast cancer more efficiently than ever before by presenting an advanced system that can be used in future researches or practical applications within medical diagnostics field with improved precision levels of diagnosis and efficiency over some current systems in place in this area of research.