Early detection of breast cancer is very crucial for effective treatment planning and improved health of patients. U-Net, a widely used deep learning model, facilitates medical image segmentation by generating lesion masks quickly, though with limited precision. Integrating deep learning with the Active Contour (AC) method enhances segmentation accuracy. This approach balances the efficiency of deep learning with the interpretability of AC. The trained U-Net model produces an initial rough tumor mask, which AC then refines by precisely attaching tumor boundaries within the edge-enhanced mask. This hybrid method proves effective in segmenting complex ultrasound images, demonstrating robust performance on diverse tumor shapes. Experimental evaluations highlight the advantages of this approach, showing improved convergence and segmentation accuracy compared to standalone U-Net models. The algorithm operates without manual tuning, though fine-tuning can further optimize performance. Quantitative analysis confirms a significant enhancement in segmentation accuracy, reinforcing the potential of this combined approach in medical imaging.