Breast cancer remains the leading cause of mortality among women worldwide, with ultrasound imaging being a prevalent method for detecting breast abnormalities. Tumor segmentation in ultrasound images poses significant challenges due to factors such as low contrast, high shadowing, and poorly defined borders. In this study, we propose an enhanced approach for breast ultrasound image segmentation by integrating a comprehensive image preprocessing pipeline with a deep learning model based on SegNet, augmented with transfer learning from MobileNet. The preprocessing phase utilized techniques including Contrast Limited Adaptive Histogram Equalization (CLAHE), Gaussian blur, and Gabor filtering to mitigate issues like acoustic shadowing, speckle noise, and low contrast, thereby enhancing the quality of the dataset, which consists of 647 breast ultrasound images (BUSI dataset). Additionally, data augmentation techniques were applied to further expand the dataset and improve model robustness. Following preprocessing and augmentation, the SegNet model with MobileNet transfer learning was trained and evaluated. Performance was assessed using metrics such as Intersection over Union (IoU), Precision, Dice Coefficient, and Accuracy. Our proposed method was also compared against four recent fully automatic segmentation techniques on the same dataset, demonstrating superior performance across.